lix/src/libcmd/installables.cc

852 lines
28 KiB
C++
Raw Normal View History

2022-03-26 10:32:38 +00:00
#include "globals.hh"
2019-10-14 12:40:16 +00:00
#include "installables.hh"
#include "installable-derived-path.hh"
#include "installable-attr-path.hh"
#include "installable-flake.hh"
#include "logging.hh"
#include "outputs-spec.hh"
2017-04-25 10:06:32 +00:00
#include "command.hh"
#include "attr-path.hh"
#include "common-eval-args.hh"
#include "derivations.hh"
#include "eval.hh"
#include "eval-settings.hh"
#include "store-api.hh"
#include "shared.hh"
#include "flake/flake.hh"
2020-04-20 11:14:59 +00:00
#include "eval-cache.hh"
#include "registry.hh"
#include "build-result.hh"
#include <nlohmann/json.hpp>
namespace nix {
void completeFlakeInputPath(
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
AddCompletions & completions,
ref<EvalState> evalState,
const std::vector<FlakeRef> & flakeRefs,
std::string_view prefix)
{
for (auto & flakeRef : flakeRefs) {
auto flake = flake::getFlake(*evalState, flakeRef, true);
for (auto & input : flake.inputs)
if (input.first.starts_with(prefix))
completions.add(input.first);
}
}
MixFlakeOptions::MixFlakeOptions()
{
2021-01-25 18:03:13 +00:00
auto category = "Common flake-related options";
addFlag({
.longName = "no-update-lock-file",
.description = "Do not allow any updates to the flake's lock file.",
2021-01-25 18:03:13 +00:00
.category = category,
.handler = {&lockFlags.updateLockFile, false}
});
addFlag({
.longName = "no-write-lock-file",
.description = "Do not write the flake's newly generated lock file.",
2021-01-25 18:03:13 +00:00
.category = category,
.handler = {&lockFlags.writeLockFile, false}
});
addFlag({
.longName = "no-registries",
2021-07-21 12:27:37 +00:00
.description =
"Don't allow lookups in the flake registries. This option is deprecated; use `--no-use-registries`.",
2021-01-25 18:03:13 +00:00
.category = category,
.handler = {[&]() {
lockFlags.useRegistries = false;
2021-07-21 12:27:37 +00:00
warn("'--no-registries' is deprecated; use '--no-use-registries'");
}}
});
addFlag({
.longName = "commit-lock-file",
.description = "Commit changes to the flake's lock file.",
2021-01-25 18:03:13 +00:00
.category = category,
.handler = {&lockFlags.commitLockFile, true}
});
2020-02-05 13:48:49 +00:00
addFlag({
.longName = "override-input",
.description = "Override a specific flake input (e.g. `dwarffs/nixpkgs`). This implies `--no-write-lock-file`.",
2021-01-25 18:03:13 +00:00
.category = category,
.labels = {"input-path", "flake-url"},
.handler = {[&](std::string inputPath, std::string flakeRef) {
lockFlags.writeLockFile = false;
lockFlags.inputOverrides.insert_or_assign(
flake::parseInputPath(inputPath),
parseFlakeRef(flakeRef, absPath("."), true));
}},
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
.completer = {[&](AddCompletions & completions, size_t n, std::string_view prefix) {
if (n == 0) {
completeFlakeInputPath(completions, getEvalState(), getFlakeRefsForCompletion(), prefix);
} else if (n == 1) {
completeFlakeRef(completions, getEvalState()->store, prefix);
}
}}
});
addFlag({
.longName = "reference-lock-file",
.description = "Read the given lock file instead of `flake.lock` within the top-level flake.",
.category = category,
.labels = {"flake-lock-path"},
.handler = {[&](std::string lockFilePath) {
lockFlags.referenceLockFilePath = lockFilePath;
}},
.completer = completePath
});
addFlag({
.longName = "output-lock-file",
.description = "Write the given lock file instead of `flake.lock` within the top-level flake.",
.category = category,
.labels = {"flake-lock-path"},
.handler = {[&](std::string lockFilePath) {
lockFlags.outputLockFilePath = lockFilePath;
}},
.completer = completePath
});
addFlag({
.longName = "inputs-from",
.description = "Use the inputs of the specified flake as registry entries.",
2021-01-25 18:03:13 +00:00
.category = category,
.labels = {"flake-url"},
.handler = {[&](std::string flakeRef) {
auto evalState = getEvalState();
auto flake = flake::lockFlake(
*evalState,
parseFlakeRef(flakeRef, absPath(".")),
{ .writeLockFile = false });
for (auto & [inputName, input] : flake.lockFile.root->inputs) {
auto input2 = flake.lockFile.findInput({inputName}); // resolve 'follows' nodes
if (auto input3 = std::dynamic_pointer_cast<const flake::LockedNode>(input2)) {
overrideRegistry(
fetchers::Input::fromAttrs({{"type","indirect"}, {"id", inputName}}),
input3->lockedRef.input,
{});
}
}
}},
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
.completer = {[&](AddCompletions & completions, size_t, std::string_view prefix) {
completeFlakeRef(completions, getEvalState()->store, prefix);
}}
});
}
SourceExprCommand::SourceExprCommand()
{
2020-05-04 20:40:19 +00:00
addFlag({
.longName = "file",
.shortName = 'f',
.description =
"Interpret [*installables*](@docroot@/command-ref/new-cli/nix.md#installables) as attribute paths relative to the Nix expression stored in *file*. "
"If *file* is the character -, then a Nix expression will be read from standard input. "
"Implies `--impure`.",
2021-01-25 18:03:13 +00:00
.category = installablesCategory,
2020-05-04 20:40:19 +00:00
.labels = {"file"},
2020-05-10 19:35:07 +00:00
.handler = {&file},
.completer = completePath
2020-05-04 20:40:19 +00:00
});
addFlag({
.longName = "expr",
.shortName = 'E',
.description = "Interpret [*installables*](@docroot@/command-ref/new-cli/nix.md#installables) as attribute paths relative to the Nix expression *expr*.",
2021-01-25 18:03:13 +00:00
.category = installablesCategory,
.labels = {"expr"},
.handler = {&expr}
});
}
MixReadOnlyOption::MixReadOnlyOption()
{
addFlag({
.longName = "read-only",
.description =
"Do not instantiate each evaluated derivation. "
"This improves performance, but can cause errors when accessing "
"store paths of derivations during evaluation.",
.handler = {&settings.readOnlyMode, true},
});
}
Strings SourceExprCommand::getDefaultFlakeAttrPaths()
{
return {
"packages." + settings.thisSystem.get() + ".default",
"defaultPackage." + settings.thisSystem.get()
};
}
Strings SourceExprCommand::getDefaultFlakeAttrPathPrefixes()
{
return {
// As a convenience, look for the attribute in
// 'outputs.packages'.
"packages." + settings.thisSystem.get() + ".",
// As a temporary hack until Nixpkgs is properly converted
// to provide a clean 'packages' set, look in 'legacyPackages'.
"legacyPackages." + settings.thisSystem.get() + "."
};
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
Args::CompleterClosure SourceExprCommand::getCompleteInstallable()
{
return [this](AddCompletions & completions, size_t, std::string_view prefix) {
completeInstallable(completions, prefix);
};
}
void SourceExprCommand::completeInstallable(AddCompletions & completions, std::string_view prefix)
{
try {
if (file) {
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
completions.setType(AddCompletions::Type::Attrs);
evalSettings.pureEval = false;
auto state = getEvalState();
Expr & e = state->parseExprFromFile(
resolveExprPath(state->checkSourcePath(lookupFileArg(*state, *file)))
);
Value root;
state->eval(e, root);
auto autoArgs = getAutoArgs(*state);
std::string prefix_ = std::string(prefix);
auto sep = prefix_.rfind('.');
std::string searchWord;
if (sep != std::string::npos) {
searchWord = prefix_.substr(sep + 1, std::string::npos);
prefix_ = prefix_.substr(0, sep);
} else {
searchWord = prefix_;
prefix_ = "";
}
auto [v, pos] = findAlongAttrPath(*state, prefix_, *autoArgs, root);
Value &v1(*v);
state->forceValue(v1, pos);
Value v2;
state->autoCallFunction(*autoArgs, v1, v2);
if (v2.type() == nAttrs) {
for (auto & i : *v2.attrs) {
std::string name = state->symbols[i.name];
if (name.find(searchWord) == 0) {
if (prefix_ == "")
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
completions.add(name);
else
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
completions.add(prefix_ + "." + name);
}
}
}
} else {
completeFlakeRefWithFragment(
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
completions,
getEvalState(),
lockFlags,
getDefaultFlakeAttrPathPrefixes(),
getDefaultFlakeAttrPaths(),
prefix);
}
} catch (EvalError&) {
// Don't want eval errors to mess-up with the completion engine, so let's just swallow them
}
2020-06-05 12:09:12 +00:00
}
void completeFlakeRefWithFragment(
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
AddCompletions & completions,
2020-06-05 12:09:12 +00:00
ref<EvalState> evalState,
flake::LockFlags lockFlags,
Strings attrPathPrefixes,
const Strings & defaultFlakeAttrPaths,
std::string_view prefix)
{
/* Look for flake output attributes that match the
prefix. */
try {
auto hash = prefix.find('#');
if (hash == std::string::npos) {
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
completeFlakeRef(completions, evalState->store, prefix);
} else {
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
completions.setType(AddCompletions::Type::Attrs);
auto fragment = prefix.substr(hash + 1);
std::string prefixRoot = "";
if (fragment.starts_with(".")){
fragment = fragment.substr(1);
prefixRoot = ".";
}
auto flakeRefS = std::string(prefix.substr(0, hash));
auto flakeRef = parseFlakeRef(expandTilde(flakeRefS), absPath("."));
2020-06-05 12:09:12 +00:00
auto evalCache = openEvalCache(*evalState,
2020-08-07 12:13:24 +00:00
std::make_shared<flake::LockedFlake>(lockFlake(*evalState, flakeRef, lockFlags)));
auto root = evalCache->getRoot();
if (prefixRoot == "."){
attrPathPrefixes.clear();
}
/* Complete 'fragment' relative to all the
attrpath prefixes as well as the root of the
flake. */
attrPathPrefixes.push_back("");
for (auto & attrPathPrefixS : attrPathPrefixes) {
2020-06-05 12:09:12 +00:00
auto attrPathPrefix = parseAttrPath(*evalState, attrPathPrefixS);
auto attrPathS = attrPathPrefixS + std::string(fragment);
2020-06-05 12:09:12 +00:00
auto attrPath = parseAttrPath(*evalState, attrPathS);
std::string lastAttr;
if (!attrPath.empty() && !attrPathS.ends_with(".")) {
lastAttr = evalState->symbols[attrPath.back()];
attrPath.pop_back();
}
auto attr = root->findAlongAttrPath(attrPath);
if (!attr) continue;
for (auto & attr2 : (*attr)->getAttrs()) {
if (std::string_view(evalState->symbols[attr2]).starts_with(lastAttr)) {
auto attrPath2 = (*attr)->getAttrPath(attr2);
/* Strip the attrpath prefix. */
attrPath2.erase(attrPath2.begin(), attrPath2.begin() + attrPathPrefix.size());
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
completions.add(flakeRefS + "#" + prefixRoot + concatStringsSep(".", evalState->symbols.resolve(attrPath2)));
}
}
}
/* And add an empty completion for the default
attrpaths. */
if (fragment.empty()) {
2020-06-05 12:09:12 +00:00
for (auto & attrPath : defaultFlakeAttrPaths) {
auto attr = root->findAlongAttrPath(parseAttrPath(*evalState, attrPath));
if (!attr) continue;
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
completions.add(flakeRefS + "#" + prefixRoot);
}
}
}
} catch (Error & e) {
warn(e.msg());
}
2020-05-11 20:10:33 +00:00
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
void completeFlakeRef(AddCompletions & completions, ref<Store> store, std::string_view prefix)
2020-05-11 20:10:33 +00:00
{
if (!experimentalFeatureSettings.isEnabled(Xp::Flakes))
return;
2020-05-11 20:10:33 +00:00
if (prefix == "")
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
completions.add(".");
2020-05-11 20:10:33 +00:00
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
Args::completeDir(completions, 0, prefix);
2020-05-11 20:10:33 +00:00
/* Look for registry entries that match the prefix. */
2020-06-05 12:09:12 +00:00
for (auto & registry : fetchers::getRegistries(store)) {
for (auto & entry : registry->entries) {
auto from = entry.from.to_string();
if (!prefix.starts_with("flake:") && from.starts_with("flake:")) {
std::string from2(from, 6);
if (from2.starts_with(prefix))
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
completions.add(from2);
} else {
if (from.starts_with(prefix))
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
completions.add(from);
}
}
}
}
DerivedPathWithInfo Installable::toDerivedPath()
2017-09-06 14:03:22 +00:00
{
auto buildables = toDerivedPaths();
2017-09-06 14:03:22 +00:00
if (buildables.size() != 1)
throw Error("installable '%s' evaluates to %d derivations, where only one is expected", what(), buildables.size());
return std::move(buildables[0]);
}
static StorePath getDeriver(
ref<Store> store,
const Installable & i,
const StorePath & drvPath)
{
auto derivers = store->queryValidDerivers(drvPath);
if (derivers.empty())
throw Error("'%s' does not have a known deriver", i.what());
// FIXME: use all derivers?
return *derivers.begin();
}
2020-04-20 11:13:52 +00:00
ref<eval_cache::EvalCache> openEvalCache(
EvalState & state,
2020-08-07 12:13:24 +00:00
std::shared_ptr<flake::LockedFlake> lockedFlake)
2020-04-20 11:13:52 +00:00
{
2020-08-04 03:46:28 +00:00
auto fingerprint = lockedFlake->getFingerprint();
return make_ref<nix::eval_cache::EvalCache>(
2020-08-07 12:13:24 +00:00
evalSettings.useEvalCache && evalSettings.pureEval
? std::optional { std::cref(fingerprint) }
: std::nullopt,
2020-04-20 11:13:52 +00:00
state,
[&state, lockedFlake]()
2020-04-20 11:13:52 +00:00
{
/* For testing whether the evaluation cache is
complete. */
if (getEnv("NIX_ALLOW_EVAL").value_or("1") == "0")
throw Error("not everything is cached, but evaluation is not allowed");
auto vFlake = state.allocValue();
flake::callFlake(state, *lockedFlake, *vFlake);
2020-04-20 11:13:52 +00:00
state.forceAttrs(*vFlake, noPos, "while parsing cached flake data");
2020-04-20 11:13:52 +00:00
auto aOutputs = vFlake->attrs->get(state.symbols.create("outputs"));
assert(aOutputs);
return aOutputs->value;
});
2020-04-20 11:13:52 +00:00
}
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
Installables SourceExprCommand::parseInstallables(
2019-04-08 14:11:17 +00:00
ref<Store> store, std::vector<std::string> ss)
{
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
Installables result;
if (file || expr) {
if (file && expr)
throw UsageError("'--file' and '--expr' are exclusive");
2019-04-08 14:11:17 +00:00
// FIXME: backward compatibility hack
if (file) evalSettings.pureEval = false;
2019-04-08 14:11:17 +00:00
auto state = getEvalState();
auto vFile = state->allocValue();
if (file == "-") {
auto & e = state->parseStdin();
state->eval(e, *vFile);
}
else if (file)
state->evalFile(lookupFileArg(*state, *file), *vFile);
else {
auto & e = state->parseExprFromString(*expr, state->rootPath(CanonPath::fromCwd()));
state->eval(e, *vFile);
}
for (auto & s : ss) {
auto [prefix, extendedOutputsSpec] = ExtendedOutputsSpec::parse(s);
result.push_back(
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
make_ref<InstallableAttrPath>(
InstallableAttrPath::parse(
state, *this, vFile, std::move(prefix), std::move(extendedOutputsSpec))));
}
2019-04-08 14:11:17 +00:00
} else {
2019-04-08 14:11:17 +00:00
for (auto & s : ss) {
std::exception_ptr ex;
auto [prefix_, extendedOutputsSpec_] = ExtendedOutputsSpec::parse(s);
// To avoid clang's pedantry
auto prefix = std::move(prefix_);
auto extendedOutputsSpec = std::move(extendedOutputsSpec_);
if (prefix.find('/') != std::string::npos) {
try {
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
result.push_back(make_ref<InstallableDerivedPath>(
InstallableDerivedPath::parse(store, prefix, extendedOutputsSpec.raw)));
continue;
} catch (BadStorePath &) {
} catch (...) {
if (!ex)
ex = std::current_exception();
}
}
try {
auto [flakeRef, fragment] = parseFlakeRefWithFragment(std::string { prefix }, absPath("."));
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
result.push_back(make_ref<InstallableFlake>(
this,
getEvalState(),
std::move(flakeRef),
fragment,
std::move(extendedOutputsSpec),
getDefaultFlakeAttrPaths(),
getDefaultFlakeAttrPathPrefixes(),
lockFlags));
continue;
} catch (...) {
ex = std::current_exception();
}
std::rethrow_exception(ex);
2019-04-08 14:11:17 +00:00
}
}
return result;
}
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
ref<Installable> SourceExprCommand::parseInstallable(
2019-04-08 14:11:17 +00:00
ref<Store> store, const std::string & installable)
{
2019-04-08 14:11:17 +00:00
auto installables = parseInstallables(store, {installable});
assert(installables.size() == 1);
return installables.front();
}
static SingleBuiltPath getBuiltPath(ref<Store> evalStore, ref<Store> store, const SingleDerivedPath & b)
{
return std::visit(
overloaded{
[&](const SingleDerivedPath::Opaque & bo) -> SingleBuiltPath {
return SingleBuiltPath::Opaque { bo.path };
},
[&](const SingleDerivedPath::Built & bfd) -> SingleBuiltPath {
auto drvPath = getBuiltPath(evalStore, store, *bfd.drvPath);
// Resolving this instead of `bfd` will yield the same result, but avoid duplicative work.
SingleDerivedPath::Built truncatedBfd {
.drvPath = makeConstantStorePathRef(drvPath.outPath()),
.output = bfd.output,
};
auto outputPath = resolveDerivedPath(*store, truncatedBfd, &*evalStore);
return SingleBuiltPath::Built {
.drvPath = make_ref<SingleBuiltPath>(std::move(drvPath)),
.output = { bfd.output, outputPath },
};
},
},
b.raw());
}
std::vector<BuiltPathWithResult> Installable::build(
ref<Store> evalStore,
ref<Store> store,
Realise mode,
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
const Installables & installables,
BuildMode bMode)
{
std::vector<BuiltPathWithResult> res;
for (auto & [_, builtPathWithResult] : build2(evalStore, store, mode, installables, bMode))
res.push_back(builtPathWithResult);
return res;
}
static void throwBuildErrors(
std::vector<KeyedBuildResult> & buildResults,
const Store & store)
{
std::vector<KeyedBuildResult> failed;
for (auto & buildResult : buildResults) {
if (!buildResult.success()) {
failed.push_back(buildResult);
}
}
auto failedResult = failed.begin();
if (failedResult != failed.end()) {
if (failed.size() == 1) {
failedResult->rethrow();
} else {
StringSet failedPaths;
for (; failedResult != failed.end(); failedResult++) {
if (!failedResult->errorMsg.empty()) {
logError(ErrorInfo{
.level = lvlError,
.msg = failedResult->errorMsg,
});
}
failedPaths.insert(failedResult->path.to_string(store));
}
throw Error("build of %s failed", concatStringsSep(", ", quoteStrings(failedPaths)));
}
}
}
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
std::vector<std::pair<ref<Installable>, BuiltPathWithResult>> Installable::build2(
2021-09-10 08:39:39 +00:00
ref<Store> evalStore,
ref<Store> store,
Realise mode,
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
const Installables & installables,
2021-09-10 08:39:39 +00:00
BuildMode bMode)
{
2020-07-15 18:05:42 +00:00
if (mode == Realise::Nothing)
settings.readOnlyMode = true;
struct Aux
{
ref<ExtraPathInfo> info;
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
ref<Installable> installable;
};
2021-04-05 13:48:18 +00:00
std::vector<DerivedPath> pathsToBuild;
std::map<DerivedPath, std::vector<Aux>> backmap;
2017-09-06 14:03:22 +00:00
for (auto & i : installables) {
for (auto b : i->toDerivedPaths()) {
pathsToBuild.push_back(b.path);
backmap[b.path].push_back({.info = b.info, .installable = i});
}
2017-09-06 14:03:22 +00:00
}
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
std::vector<std::pair<ref<Installable>, BuiltPathWithResult>> res;
switch (mode) {
case Realise::Nothing:
case Realise::Derivation:
2017-09-06 14:03:22 +00:00
printMissing(store, pathsToBuild, lvlError);
for (auto & path : pathsToBuild) {
for (auto & aux : backmap[path]) {
std::visit(overloaded {
[&](const DerivedPath::Built & bfd) {
auto outputs = resolveDerivedPath(*store, bfd, &*evalStore);
res.push_back({aux.installable, {
.path = BuiltPath::Built {
.drvPath = make_ref<SingleBuiltPath>(getBuiltPath(evalStore, store, *bfd.drvPath)),
.outputs = outputs,
},
.info = aux.info}});
},
[&](const DerivedPath::Opaque & bo) {
res.push_back({aux.installable, {
.path = BuiltPath::Opaque { bo.path },
.info = aux.info}});
},
}, path.raw());
}
}
break;
case Realise::Outputs: {
if (settings.printMissing)
2022-11-18 12:40:48 +00:00
printMissing(store, pathsToBuild, lvlInfo);
auto buildResults = store->buildPathsWithResults(pathsToBuild, bMode, evalStore);
throwBuildErrors(buildResults, *store);
for (auto & buildResult : buildResults) {
for (auto & aux : backmap[buildResult.path]) {
std::visit(overloaded {
[&](const DerivedPath::Built & bfd) {
std::map<std::string, StorePath> outputs;
for (auto & [outputName, realisation] : buildResult.builtOutputs)
outputs.emplace(outputName, realisation.outPath);
res.push_back({aux.installable, {
.path = BuiltPath::Built {
.drvPath = make_ref<SingleBuiltPath>(getBuiltPath(evalStore, store, *bfd.drvPath)),
.outputs = outputs,
},
.info = aux.info,
.result = buildResult}});
},
[&](const DerivedPath::Opaque & bo) {
res.push_back({aux.installable, {
.path = BuiltPath::Opaque { bo.path },
.info = aux.info,
.result = buildResult}});
},
}, buildResult.path.raw());
}
}
break;
}
default:
assert(false);
}
return res;
2017-09-06 14:03:22 +00:00
}
2022-03-02 12:54:08 +00:00
BuiltPaths Installable::toBuiltPaths(
ref<Store> evalStore,
ref<Store> store,
Realise mode,
OperateOn operateOn,
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
const Installables & installables)
2017-09-06 14:03:22 +00:00
{
if (operateOn == OperateOn::Output) {
BuiltPaths res;
for (auto & p : Installable::build(evalStore, store, mode, installables))
res.push_back(p.path);
return res;
} else {
if (mode == Realise::Nothing)
settings.readOnlyMode = true;
BuiltPaths res;
2022-03-02 12:54:08 +00:00
for (auto & drvPath : Installable::toDerivations(store, installables, true))
res.emplace_back(BuiltPath::Opaque{drvPath});
return res;
}
}
StorePathSet Installable::toStorePathSet(
ref<Store> evalStore,
ref<Store> store,
Realise mode, OperateOn operateOn,
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
const Installables & installables)
{
StorePathSet outPaths;
for (auto & path : toBuiltPaths(evalStore, store, mode, operateOn, installables)) {
auto thisOutPaths = path.outPaths();
outPaths.insert(thisOutPaths.begin(), thisOutPaths.end());
}
return outPaths;
}
StorePaths Installable::toStorePaths(
ref<Store> evalStore,
ref<Store> store,
Realise mode, OperateOn operateOn,
const Installables & installables)
{
StorePaths outPaths;
for (auto & path : toBuiltPaths(evalStore, store, mode, operateOn, installables)) {
auto thisOutPaths = path.outPaths();
outPaths.insert(outPaths.end(), thisOutPaths.begin(), thisOutPaths.end());
}
return outPaths;
}
2022-03-02 12:54:08 +00:00
StorePath Installable::toStorePath(
ref<Store> evalStore,
ref<Store> store,
Realise mode, OperateOn operateOn,
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
ref<Installable> installable)
{
auto paths = toStorePathSet(evalStore, store, mode, operateOn, {installable});
if (paths.size() != 1)
2019-04-08 21:58:33 +00:00
throw Error("argument '%s' should evaluate to one store path", installable->what());
return *paths.begin();
}
2022-03-02 12:54:08 +00:00
StorePathSet Installable::toDerivations(
2021-09-10 08:39:39 +00:00
ref<Store> store,
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
const Installables & installables,
2021-09-10 08:39:39 +00:00
bool useDeriver)
{
StorePathSet drvPaths;
for (const auto & i : installables)
for (const auto & b : i->toDerivedPaths())
std::visit(overloaded {
[&](const DerivedPath::Opaque & bo) {
Get rid of `.drv` special-casing for store path installables The release notes document the change in behavior, I don't include it here so there is no risk to it getting out of sync. > Motivation >> Plumbing CLI should be simple Store derivation installations are intended as "plumbing": very simple utilities for advanced users and scripts, and not what regular users interact with. (Similarly, regular Git users will use branch and tag names not explicit hashes for most things.) The plumbing CLI should prize simplicity over convenience; that is its raison d'etre. If the user provides a path, we should treat it the same way not caring what sort of path it is. >> Scripting This is especially important for the scripting use-case. when arbitrary paths are sent to e.g. `nix copy` and the script author wants consistent behavior regardless of what those store paths are. Otherwise the script author needs to be careful to filter out `.drv` ones, and then run `nix copy` again with those paths and `--derivation`. That is not good! >> Surprisingly low impact Only two lines in the tests need changing, showing that the impact of this is pretty light. Many command, like `nix log` will continue to work with just the derivation passed as before. This because we used to: - Special case the drv path and replace it with it's outputs (what this gets rid of). - Turn those output path *back* into the original drv path. Now we just skip that entire round trip! > Context Issue #7261 lays out a broader vision for getting rid of `--derivation`, and has this as one of its dependencies. But we can do this with or without that. `Installable::toDerivations` is changed to handle the case of a `DerivedPath::Opaque` ending in `.drv`, which is new: it simply doesn't need to do any extra work in that case. On this basis, commands like `nix {show-derivation,log} /nix/store/...-foo.drv` still work as before, as described above. When testing older daemons, the post-build-hook will be run against the old CLI, so we need the old version of the post-build-hook to support that use-case. Co-authored-by: Travis A. Everett <travis.a.everett@gmail.com> Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io>
2022-12-16 03:44:14 +00:00
drvPaths.insert(
bo.path.isDerivation()
? bo.path
: useDeriver
? getDeriver(store, *i, bo.path)
: throw Error("argument '%s' did not evaluate to a derivation", i->what()));
},
[&](const DerivedPath::Built & bfd) {
drvPaths.insert(resolveDerivedPath(*store, *bfd.drvPath));
},
}, b.path.raw());
return drvPaths;
}
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
RawInstallablesCommand::RawInstallablesCommand()
2020-05-11 13:46:18 +00:00
{
addFlag({
.longName = "stdin",
.description = "Read installables from the standard input. No default installable applied.",
.handler = {&readFromStdIn, true}
});
2020-05-11 13:46:18 +00:00
expectArgs({
.label = "installables",
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
.handler = {&rawInstallables},
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
.completer = getCompleteInstallable(),
2020-05-11 13:46:18 +00:00
});
}
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
void RawInstallablesCommand::applyDefaultInstallables(std::vector<std::string> & rawInstallables)
{
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
if (rawInstallables.empty()) {
2022-03-26 10:32:37 +00:00
// FIXME: commands like "nix profile install" should not have a
// default, probably.
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
rawInstallables.push_back(".");
}
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
std::vector<FlakeRef> RawInstallablesCommand::getFlakeRefsForCompletion()
{
applyDefaultInstallables(rawInstallables);
std::vector<FlakeRef> res;
for (auto i : rawInstallables)
res.push_back(parseFlakeRefWithFragment(
expandTilde(i),
absPath(".")).first);
return res;
}
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
void RawInstallablesCommand::run(ref<Store> store)
{
if (readFromStdIn && !isatty(STDIN_FILENO)) {
std::string word;
while (std::cin >> word) {
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
rawInstallables.emplace_back(std::move(word));
}
} else {
applyDefaultInstallables(rawInstallables);
}
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
run(store, std::move(rawInstallables));
}
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
std::vector<FlakeRef> InstallableCommand::getFlakeRefsForCompletion()
2020-06-08 14:20:00 +00:00
{
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
return {
parseFlakeRefWithFragment(
expandTilde(_installable),
absPath(".")).first
};
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
}
void InstallablesCommand::run(ref<Store> store, std::vector<std::string> && rawInstallables)
{
auto installables = parseInstallables(store, rawInstallables);
run(store, std::move(installables));
2020-06-08 14:20:00 +00:00
}
InstallableCommand::InstallableCommand()
: SourceExprCommand()
{
expectArgs({
.label = "installable",
2020-05-12 09:53:32 +00:00
.optional = true,
.handler = {&_installable},
Overhaul completions, redo #6693 (#8131) As I complained in https://github.com/NixOS/nix/pull/6784#issuecomment-1421777030 (a comment on the wrong PR, sorry again!), #6693 introduced a second completions mechanism to fix a bug. Having two completion mechanisms isn't so nice. As @thufschmitt also pointed out, it was a bummer to go from `FlakeRef` to `std::string` when collecting flake refs. Now it is `FlakeRefs` again. The underlying issue that sought to work around was that completion of arguments not at the end can still benefit from the information from latter arguments. To fix this better, we rip out that change and simply defer all completion processing until after all the (regular, already-complete) arguments have been passed. In addition, I noticed the original completion logic used some global variables. I do not like global variables, because even if they save lines of code, they also obfuscate the architecture of the code. I got rid of them moved them to a new `RootArgs` class, which now has `parseCmdline` instead of `Args`. The idea is that we have many argument parsers from subcommands and what-not, but only one root args that owns the other per actual parsing invocation. The state that was global is now part of the root args instead. This did, admittedly, add a bunch of new code. And I do feel bad about that. So I went and added a lot of API docs to try to at least make the current state of things clear to the next person. -- This is needed for RFC 134 (tracking issue #7868). It was very hard to modularize `Installable` parsing when there were two completion arguments. I wouldn't go as far as to say it is *easy* now, but at least it is less hard (and the completions test finally passed). Co-authored-by: Valentin Gagarin <valentin.gagarin@tweag.io> Change-Id: If18cd5be78da4a70635e3fdcac6326dbfeea71a5 (cherry picked from commit 67eb37c1d0de28160cd25376e51d1ec1b1c8305b)
2024-03-19 02:23:20 +00:00
.completer = getCompleteInstallable(),
});
}
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
void InstallableCommand::run(ref<Store> store)
{
auto installable = parseInstallable(store, _installable);
run(store, std::move(installable));
}
void BuiltPathsCommand::applyDefaultInstallables(std::vector<std::string> & rawInstallables)
{
Make command infra less stateful and more regular Already, we had classes like `BuiltPathsCommand` and `StorePathsCommand` which provided alternative `run` virtual functions providing the implementation with more arguments. This was a very nice and easy way to make writing command; just fill in the virtual functions and it is fairly clear what to do. However, exception to this pattern were `Installable{,s}Command`. These two classes instead just had a field where the installables would be stored, and various side-effecting `prepare` and `load` machinery too fill them in. Command would wish out those fields. This isn't so clear to use. What this commit does is make those command classes like the others, with richer `run` functions. Not only does this restore the pattern making commands easier to write, it has a number of other benefits: - `prepare` and `load` are gone entirely! One command just hands just hands off to the next. - `useDefaultInstallables` because `defaultInstallables`. This takes over `prepare` for the one case that needs it, and provides enough flexiblity to handle `nix repl`'s idiosyncratic migration. - We can use `ref` instead of `std::shared_ptr`. The former must be initialized (so it is like Rust's `Box` rather than `Option<Box>`, This expresses the invariant that the installable are in fact initialized much better. This is possible because since we just have local variables not fields, we can stop worrying about the not-yet-initialized case. - Fewer lines of code! (Finally I have a large refactor that makes the number go down not up...) - `nix repl` is now implemented in a clearer way. The last item deserves further mention. `nix repl` is not like the other installable commands because instead working from once-loaded installables, it needs to be able to load them again and again. To properly support this, we make a new superclass `RawInstallablesCommand`. This class has the argument parsing and completion logic, but does *not* hand off parsed installables but instead just the raw string arguments. This is exactly what `nix repl` needs, and allows us to instead of having the logic awkwardly split between `prepare`, `useDefaultInstallables,` and `load`, have everything right next to each other. I think this will enable future simplifications of that argument defaulting logic, but I am saving those for a future PR --- best to keep code motion and more complicated boolean expression rewriting separate steps. The "diagnostic ignored `-Woverloaded-virtual`" pragma helps because C++ doesn't like our many `run` methods. In our case, we don't mind the shadowing it all --- it is *intentional* that the derived class only provides a `run` method, and doesn't call any of the overridden `run` methods. Helps with https://github.com/NixOS/rfcs/pull/134
2023-02-04 17:03:47 +00:00
if (rawInstallables.empty() && !all)
rawInstallables.push_back(".");
}
}