Motivation
`PathSet` is not correct because string contexts have other forms
(`Built` and `DrvDeep`) that are not rendered as plain store paths.
Instead of wrongly using `PathSet`, or "stringly typed" using
`StringSet`, use `std::std<StringContextElem>`.
-----
In support of this change, `NixStringContext` is now defined as
`std::std<StringContextElem>` not `std:vector<StringContextElem>`. The
old definition was just used by a `getContext` method which was only
used by the eval cache. It can be deleted altogether since the types are
now unified and the preexisting `copyContext` function already suffices.
Summarizing the previous paragraph:
Old:
- `value/context.hh`: `NixStringContext = std::vector<StringContextElem>`
- `value.hh`: `NixStringContext Value::getContext(...)`
- `value.hh`: `copyContext(...)`
New:
- `value/context.hh`: `NixStringContext = std::set<StringContextElem>`
- `value.hh`: `copyContext(...)`
----
The string representation of string context elements no longer contains
the store dir. The diff of `src/libexpr/tests/value/context.cc` should
make clear what the new representation is, so we recommend reviewing
that file first. This was done for two reasons:
Less API churn:
`Value::mkString` and friends did not take a `Store` before. But if
`NixStringContextElem::{parse, to_string}` *do* take a store (as they
did before), then we cannot have the `Value` functions use them (in
order to work with the fully-structured `NixStringContext`) without
adding that argument.
That would have been a lot of churn of threading the store, and this
diff is already large enough, so the easier and less invasive thing to
do was simply make the element `parse` and `to_string` functions not
take the `Store` reference, and the easiest way to do that was to simply
drop the store dir.
Space usage:
Dropping the `/nix/store/` (or similar) from the internal representation
will safe space in the heap of the Nix programming being interpreted. If
the heap contains many strings with non-trivial contexts, the saving
could add up to something significant.
----
The eval cache version is bumped.
The eval cache serialization uses `NixStringContextElem::{parse,
to_string}`, and since those functions are changed per the above, that
means the on-disk representation is also changed.
This is simply done by changing the name of the used for the eval cache
from `eval-cache-v4` to eval-cache-v5`.
----
To avoid some duplication `EvalCache::mkPathString` is added to abstract
over the simple case of turning a store path to a string with just that
string in the context.
Context
This PR picks up where #7543 left off. That one introduced the fully
structured `NixStringContextElem` data type, but kept `PathSet context`
as an awkward middle ground between internal `char[][]` interpreter heap
string contexts and `NixStringContext` fully parsed string contexts.
The infelicity of `PathSet context` was specifically called out during
Nix team group review, but it was agreeing that fixing it could be left
as future work. This is that future work.
A possible follow-up step would be to get rid of the `char[][]`
evaluator heap representation, too, but it is not yet clear how to do
that. To use `NixStringContextElem` there we would need to get the STL
containers to GC pointers in the GC build, and I am not sure how to do
that.
----
PR #7543 effectively is writing the inverse of a `mkPathString`,
`mkOutputString`, and one more such function for the `DrvDeep` case. I
would like that PR to have property tests ensuring it is actually the
inverse as expected.
This PR sets things up nicely so that reworking that PR to be in that
more elegant and better tested way is possible.
Co-authored-by: Théophane Hufschmitt <7226587+thufschmitt@users.noreply.github.com>
This introduces the SourcePath type from lazy-trees as an abstraction
for accessing files from inputs that may not be materialized in the
real filesystem (e.g. Git repositories). Currently, however, it's just
a wrapper around CanonPath, so it shouldn't change any behaviour. (On
lazy-trees, SourcePath is a <InputAccessor, CanonPath> tuple.)
This makes the position object used in exceptions abstract, with a
method getSource() to get the source code of the file in which the
error originated. This is needed for lazy trees because source files
don't necessarily exist in the filesystem, and we don't want to make
libutil depend on the InputAccessor type in libfetcher.
this slightly increases the amount of memory used for any given symbol, but this
increase is more than made up for if the symbol is referenced more than once in
the EvalState that holds it. on average every symbol should be referenced at
least twice (once to introduce a binding, once to use it), so we expect no
increase in memory on average.
symbol tables are limited to 2³² entries like position tables, and similar
arguments apply to why overflow is not likely: 2³² symbols would require as many
string instances (at 24 bytes each) and map entries (at 24 bytes or more each,
assuming that the map holds on average at most one item per bucket as the docs
say). a full symbol table would require at least 192GB of memory just for
symbols, which is well out of reach. (an ofborg eval of nixpks today creates
less than a million symbols!)
Pos objects are somewhat wasteful as they duplicate the origin file name and
input type for each object. on files that produce more than one Pos when parsed
this a sizeable waste of memory (one pointer per Pos). the same goes for
ptr<Pos> on 64 bit machines: parsing enough source to require 8 bytes to locate
a position would need at least 8GB of input and 64GB of expression memory. it's
not likely that we'll hit that any time soon, so we can use a uint32_t index to
locate positions instead.
the only use of this function is to determine whether a lambda has a non-set
formal, but this use is arguably better served by Symbol::set and using a
non-Symbol instead of an empty symbol in the parser when no such formal is present.
if we defer the duplicate argument check for lambda formals we can use more
efficient data structures for the formals set, and we can get rid of the
duplication of formals names to boot. instead of a list of formals we've seen
and a set of names we'll keep a vector instead and run a sort+dupcheck step
before moving the parsed formals into a newly created lambda. this improves
performance on search and rebuild by ~1%, pure parsing gains more (about 4%).
this does reorder lambda arguments in the xml output, but the output is still
stable. this shouldn't be a problem since argument order is not semantically
important anyway.
before
nix search --no-eval-cache --offline ../nixpkgs hello
Time (mean ± σ): 8.550 s ± 0.060 s [User: 6.470 s, System: 1.664 s]
Range (min … max): 8.435 s … 8.666 s 20 runs
nix eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
Time (mean ± σ): 346.7 ms ± 2.1 ms [User: 312.4 ms, System: 34.2 ms]
Range (min … max): 343.8 ms … 353.4 ms 20 runs
nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
Time (mean ± σ): 2.720 s ± 0.031 s [User: 2.415 s, System: 0.231 s]
Range (min … max): 2.662 s … 2.780 s 20 runs
after
nix search --no-eval-cache --offline ../nixpkgs hello
Time (mean ± σ): 8.462 s ± 0.063 s [User: 6.398 s, System: 1.661 s]
Range (min … max): 8.339 s … 8.542 s 20 runs
nix eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
Time (mean ± σ): 329.1 ms ± 1.4 ms [User: 296.8 ms, System: 32.3 ms]
Range (min … max): 326.1 ms … 330.8 ms 20 runs
nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
Time (mean ± σ): 2.687 s ± 0.035 s [User: 2.392 s, System: 0.228 s]
Range (min … max): 2.626 s … 2.754 s 20 runs
- This change applies to builtins.toXML and inner workings
- Proof of concept:
```nix
let e = builtins.toXML e; in e
```
- Before:
```
$ nix-instantiate --eval poc.nix
error: infinite recursion encountered
```
- After:
```
$ nix-instantiate --eval poc.nix
error: infinite recursion encountered
at /data/github/kamadorueda/nix/poc.nix:1:9:
1| let e = builtins.toXML e; in e
|
```
The boolean is only used to determine if the formals are set to a
non-null pointer in all our cases. We can get rid of that allocation and
instead just compare the pointer value with NULL. Saving up to
sizeof(bool) + platform specific alignment per ExprLambda instace.
Probably not a lot of memory but perhaps a few kilobyte with nixpkgs?
This also gets rid of a potential issue with dereferencing formals based on
the value of the boolean that didn't have to be aligned with the formals
pointer but was in all our cases.
This fixes a class of crashes and introduces ptr<T> to make the
code robust against this failure mode going forward.
Thanks regnat for the idea of a ref<T> without overhead!
Closes#4895Closes#4893Closes#5127Closes#5113
The value pointers of lists with 1 or 2 elements are now stored in the
list value itself. In particular, this makes the "concatMap (x: if
cond then [(f x)] else [])" idiom cheaper.
Code that links to libnixexpr (e.g. plugins loaded with importNative, or
nix-exec) may want to provide custom value types and operations on
values of those types. For example, nix-exec is currently using sets
where a custom IO value type would be more appropriate. This commit
provides a generic hook for such types in the form of tExternal and the
ExternalBase virtual class, which contains all functions necessary for
libnixexpr's type-polymorphic functions (e.g. `showType`) to be
implemented.
tree). This saves a lot of memory. The vector should be sorted so
that names can be looked up using binary search, but this is not the
case yet. (Surprisingly, looking up attributes using linear search
doesn't have a big impact on performance.)
Memory consumption for
$ nix-instantiate /etc/nixos/nixos/tests -A bittorrent.test --readonly-mode
on x86_64-linux with GC enabled is now 185 MiB (compared to 946
MiB on the trunk).
a pointer to a Value, rather than the Value directly. This improves
the effectiveness of garbage collection a lot: if the Value is
stored inside the set directly, then any live pointer to the Value
causes all other attributes in the set to be live as well.
values. This improves sharing and gives another speed up.
Evaluation of the NixOS system attribute is now almost 7 times
faster than the old evaluator.
efficiently. The symbol table ensures that there is only one copy
of each symbol, thus allowing symbols to be compared efficiently
using a pointer equality test.
jade
Robert Hensing