rewrite the parser with pegtl instead of flex/bison

this gives about 20% performance improvements on pure parsing. obviously
it'll be less on full eval, but depending on how much parsing has to be
done (eg whether nixpkgs haskell modules are included or not) it ranges
anywhere from 4% to 10% in our tests.

this has been tested with thousands of core hours of fuzz testing to
ensure that the ASTs produced by the new parser are exactly the same as
the ones produced by the old parser. error messages will
change (sometimes a lot) and are currently not perfect, but we'd rather
leave that open for improvement than having this work rot forever.

Change-Id: Ie66ec2d045dec964632c6541e25f8f0797319ee2
This commit is contained in:
eldritch horrors 2024-03-16 16:13:47 +01:00
parent 8225284df3
commit 112fd6c971
14 changed files with 1707 additions and 912 deletions

View file

@ -342,6 +342,16 @@ AC_SUBST(doc_generate)
# Look for lowdown library.
PKG_CHECK_MODULES([LOWDOWN], [lowdown >= 0.9.0], [CXXFLAGS="$LOWDOWN_CFLAGS $CXXFLAGS"])
# Look for pegtl.
# pegtl has only cmake support, no pkg-config.
AC_ARG_VAR([PEGTL_HEADERS], [include path of pegtl headers])
AC_LANG_PUSH(C++)
AC_SUBST(PEGTL_HEADERS)
[CXXFLAGS="-I $PEGTL_HEADERS $CXXFLAGS"]
AC_CHECK_HEADER(tao/pegtl.hpp, [], [AC_MSG_ERROR([PEGTL not found.])])
AC_LANG_POP(C++)
# Setuid installations.
AC_CHECK_FUNCS([setresuid setreuid lchown])

View file

@ -192,6 +192,25 @@
boehmgc = final.boehmgc-nix;
busybox-sandbox-shell = final.busybox-sandbox-shell or final.default-busybox-sandbox-shell;
};
pegtl = final.callPackage (
{ stdenv, cmake, ninja }:
stdenv.mkDerivation {
pname = "pegtl";
version = "3.2.7";
src = final.fetchFromGitHub {
repo = "PEGTL";
owner = "taocpp";
rev = "refs/tags/3.2.7";
hash = "sha256-IV5YNGE4EWVrmg2Sia/rcU8jCuiBynQGJM6n3DCWTQU=";
};
nativeBuildInputs = [ cmake ninja ];
}
) {};
};
in {

View file

@ -7,7 +7,6 @@
aws-sdk-cpp,
boehmgc,
nlohmann_json,
bison,
changelog-d,
boost,
brotli,
@ -16,7 +15,6 @@
doxygen,
editline,
fileset,
flex,
git,
gtest,
jq,
@ -29,6 +27,7 @@
mdbook-linkcheck,
mercurial,
openssl,
pegtl,
pkg-config,
rapidcheck,
sqlite,
@ -127,9 +126,6 @@ in stdenv.mkDerivation (finalAttrs: {
dontBuild = false;
nativeBuildInputs = [
bison
flex
] ++ [
(lib.getBin lowdown)
mdbook
mdbook-linkcheck
@ -212,7 +208,8 @@ in stdenv.mkDerivation (finalAttrs: {
++ lib.optionals (finalAttrs.doCheck || internalApiDocs) testConfigureFlags
++ lib.optional (!canRunInstalled) "--disable-doc-gen"
++ [ (lib.enableFeature internalApiDocs "internal-api-docs") ]
++ lib.optional (!forDevShell) "--sysconfdir=/etc";
++ lib.optional (!forDevShell) "--sysconfdir=/etc"
++ [ "PEGTL_HEADERS=${lib.getDev pegtl}/include" ];
installTargets = lib.optional internalApiDocs "internal-api-html";

View file

@ -18,7 +18,6 @@
#include "gc-small-vector.hh"
#include "fetch-to-store.hh"
#include "flake/flakeref.hh"
#include "parser-tab.hh"
#include <algorithm>
#include <chrono>
@ -2858,21 +2857,6 @@ std::optional<std::string> EvalState::resolveSearchPathPath(const SearchPath::Pa
}
Expr * EvalState::parse(
char * text,
size_t length,
Pos::Origin origin,
const SourcePath & basePath,
std::shared_ptr<StaticEnv> & staticEnv)
{
auto result = parseExprFromBuf(text, length, origin, basePath, symbols, positions, exprSymbols);
result->bindVars(*this, staticEnv);
return result;
}
std::string ExternalValueBase::coerceToString(EvalState & state, const PosIdx & pos, NixStringContext & context, bool copyMore, bool copyToStore) const
{
state.error<TypeError>(

View file

@ -1,302 +0,0 @@
%option reentrant bison-bridge bison-locations
%option align
%option noyywrap
%option never-interactive
%option stack
%option nodefault
%option nounput noyy_top_state
%s DEFAULT
%x STRING
%x IND_STRING
%x INPATH
%x INPATH_SLASH
%x PATH_START
%{
#ifdef __clang__
#pragma clang diagnostic ignored "-Wunneeded-internal-declaration"
#endif
// yacc generates code that uses unannotated fallthrough.
#pragma GCC diagnostic ignored "-Wimplicit-fallthrough"
#ifdef __clang__
#pragma clang diagnostic ignored "-Wimplicit-fallthrough"
#endif
#include <boost/lexical_cast.hpp>
#include "nixexpr.hh"
#include "parser-tab.hh"
using namespace nix;
#define THROW(...) \
do { \
state->error.reset(new auto(__VA_ARGS__)); \
return YYerror; \
} while (0)
namespace nix {
#define CUR_POS state->at(*yylloc)
static void initLoc(YYLTYPE * loc)
{
loc->first_line = loc->last_line = 0;
loc->first_column = loc->last_column = 0;
}
static void adjustLoc(YYLTYPE * loc, const char * s, size_t len)
{
loc->stash();
loc->first_column = loc->last_column;
loc->last_column += len;
}
// we make use of the fact that the parser receives a private copy of the input
// string and can munge around in it.
static StringToken unescapeStr(SymbolTable & symbols, char * s, size_t length)
{
char * result = s;
char * t = s;
char c;
// the input string is terminated with *two* NULs, so we can safely take
// *one* character after the one being checked against.
while ((c = *s++)) {
if (c == '\\') {
c = *s++;
if (c == 'n') *t = '\n';
else if (c == 'r') *t = '\r';
else if (c == 't') *t = '\t';
else *t = c;
}
else if (c == '\r') {
/* Normalise CR and CR/LF into LF. */
*t = '\n';
if (*s == '\n') s++; /* cr/lf */
}
else *t = c;
t++;
}
return {result, size_t(t - result)};
}
}
#define YY_USER_INIT initLoc(yylloc)
#define YY_USER_ACTION adjustLoc(yylloc, yytext, yyleng);
#define PUSH_STATE(state) yy_push_state(state, yyscanner)
#define POP_STATE() yy_pop_state(yyscanner)
%}
ANY .|\n
ID [a-zA-Z\_][a-zA-Z0-9\_\'\-]*
INT [0-9]+
FLOAT (([1-9][0-9]*\.[0-9]*)|(0?\.[0-9]+))([Ee][+-]?[0-9]+)?
PATH_CHAR [a-zA-Z0-9\.\_\-\+]
PATH {PATH_CHAR}*(\/{PATH_CHAR}+)+\/?
PATH_SEG {PATH_CHAR}*\/
HPATH \~(\/{PATH_CHAR}+)+\/?
HPATH_START \~\/
SPATH \<{PATH_CHAR}+(\/{PATH_CHAR}+)*\>
URI [a-zA-Z][a-zA-Z0-9\+\-\.]*\:[a-zA-Z0-9\%\/\?\:\@\&\=\+\$\,\-\_\.\!\~\*\']+
%%
if { return IF; }
then { return THEN; }
else { return ELSE; }
assert { return ASSERT; }
with { return WITH; }
let { return LET; }
in { return IN; }
rec { return REC; }
inherit { return INHERIT; }
or { return OR_KW; }
\.\.\. { return ELLIPSIS; }
\=\= { return EQ; }
\!\= { return NEQ; }
\<\= { return LEQ; }
\>\= { return GEQ; }
\&\& { return AND; }
\|\| { return OR; }
\-\> { return IMPL; }
\/\/ { return UPDATE; }
\+\+ { return CONCAT; }
{ID} { yylval->id = {yytext, (size_t) yyleng}; return ID; }
{INT} { errno = 0;
try {
yylval->n = boost::lexical_cast<int64_t>(yytext);
} catch (const boost::bad_lexical_cast &) {
THROW(ParseError({
.msg = HintFmt("invalid integer '%1%'", yytext),
.pos = state->positions[CUR_POS],
}));
}
return INT;
}
{FLOAT} { errno = 0;
yylval->nf = strtod(yytext, 0);
if (errno != 0)
THROW(ParseError({
.msg = HintFmt("invalid float '%1%'", yytext),
.pos = state->positions[CUR_POS],
}));
return FLOAT;
}
\$\{ { PUSH_STATE(DEFAULT); return DOLLAR_CURLY; }
\} { /* State INITIAL only exists at the bottom of the stack and is
used as a marker. DEFAULT replaces it everywhere else.
Popping when in INITIAL state causes an empty stack exception,
so don't */
if (YYSTATE != INITIAL)
POP_STATE();
return '}';
}
\{ { PUSH_STATE(DEFAULT); return '{'; }
\" { PUSH_STATE(STRING); return '"'; }
<STRING>([^\$\"\\]|\$[^\{\"\\]|\\{ANY}|\$\\{ANY})*\$/\" |
<STRING>([^\$\"\\]|\$[^\{\"\\]|\\{ANY}|\$\\{ANY})+ {
/* It is impossible to match strings ending with '$' with one
regex because trailing contexts are only valid at the end
of a rule. (A sane but undocumented limitation.) */
yylval->str = unescapeStr(state->symbols, yytext, yyleng);
return STR;
}
<STRING>\$\{ { PUSH_STATE(DEFAULT); return DOLLAR_CURLY; }
<STRING>\" { POP_STATE(); return '"'; }
<STRING>\$|\\|\$\\ {
/* This can only occur when we reach EOF, otherwise the above
(...|\$[^\{\"\\]|\\.|\$\\.)+ would have triggered.
This is technically invalid, but we leave the problem to the
parser who fails with exact location. */
return EOF;
}
\'\'(\ *\n)? { PUSH_STATE(IND_STRING); return IND_STRING_OPEN; }
<IND_STRING>([^\$\']|\$[^\{\']|\'[^\'\$])+ {
yylval->str = {yytext, (size_t) yyleng, true};
return IND_STR;
}
<IND_STRING>\'\'\$ |
<IND_STRING>\$ {
yylval->str = {"$", 1};
return IND_STR;
}
<IND_STRING>\'\'\' {
yylval->str = {"''", 2};
return IND_STR;
}
<IND_STRING>\'\'\\{ANY} {
yylval->str = unescapeStr(state->symbols, yytext + 2, yyleng - 2);
return IND_STR;
}
<IND_STRING>\$\{ { PUSH_STATE(DEFAULT); return DOLLAR_CURLY; }
<IND_STRING>\'\' { POP_STATE(); return IND_STRING_CLOSE; }
<IND_STRING>\' {
yylval->str = {"'", 1};
return IND_STR;
}
{PATH_SEG}\$\{ |
{HPATH_START}\$\{ {
PUSH_STATE(PATH_START);
yyless(0);
yylloc->unstash();
}
<PATH_START>{PATH_SEG} {
POP_STATE();
PUSH_STATE(INPATH_SLASH);
yylval->path = {yytext, (size_t) yyleng};
return PATH;
}
<PATH_START>{HPATH_START} {
POP_STATE();
PUSH_STATE(INPATH_SLASH);
yylval->path = {yytext, (size_t) yyleng};
return HPATH;
}
{PATH} {
if (yytext[yyleng-1] == '/')
PUSH_STATE(INPATH_SLASH);
else
PUSH_STATE(INPATH);
yylval->path = {yytext, (size_t) yyleng};
return PATH;
}
{HPATH} {
if (yytext[yyleng-1] == '/')
PUSH_STATE(INPATH_SLASH);
else
PUSH_STATE(INPATH);
yylval->path = {yytext, (size_t) yyleng};
return HPATH;
}
<INPATH,INPATH_SLASH>\$\{ {
POP_STATE();
PUSH_STATE(INPATH);
PUSH_STATE(DEFAULT);
return DOLLAR_CURLY;
}
<INPATH,INPATH_SLASH>{PATH}|{PATH_SEG}|{PATH_CHAR}+ {
POP_STATE();
if (yytext[yyleng-1] == '/')
PUSH_STATE(INPATH_SLASH);
else
PUSH_STATE(INPATH);
yylval->str = {yytext, (size_t) yyleng};
return STR;
}
<INPATH>{ANY} |
<INPATH><<EOF>> {
/* if we encounter a non-path character we inform the parser that the path has
ended with a PATH_END token and re-parse this character in the default
context (it may be ')', ';', or something of that sort) */
POP_STATE();
yyless(0);
yylloc->unstash();
return PATH_END;
}
<INPATH_SLASH>{ANY} |
<INPATH_SLASH><<EOF>> {
THROW(ParseError({
.msg = HintFmt("path has a trailing slash"),
.pos = state->positions[CUR_POS],
}));
}
{SPATH} { yylval->path = {yytext, (size_t) yyleng}; return SPATH; }
{URI} { yylval->uri = {yytext, (size_t) yyleng}; return URI; }
[ \t\r\n]+ /* eat up whitespace */
\#[^\r\n]* /* single-line comments */
\/\*([^*]|\*+[^*/])*\*+\/ /* long comments */
{ANY} {
/* Don't return a negative number, as this will cause
Bison to stop parsing without an error. */
return (unsigned char) yytext[0];
}
%%

View file

@ -7,12 +7,12 @@ libexpr_DIR := $(d)
libexpr_SOURCES := \
$(wildcard $(d)/*.cc) \
$(wildcard $(d)/value/*.cc) \
$(wildcard $(d)/parser/*.cc) \
$(wildcard $(d)/primops/*.cc) \
$(wildcard $(d)/flake/*.cc) \
$(d)/lexer-tab.cc \
$(d)/parser-tab.cc
$(wildcard $(d)/flake/*.cc)
libexpr_CXXFLAGS += -I src/libutil -I src/libstore -I src/libfetchers -I src/libmain -I src/libexpr
libexpr_CXXFLAGS += -I src/libutil -I src/libstore -I src/libfetchers -I src/libmain -I src/libexpr \
-I pegtl/include
libexpr_LIBS = libutil libstore libfetchers
@ -26,16 +26,6 @@ endif
# because inline functions in libexpr's header files call libgc.
libexpr_LDFLAGS_PROPAGATED = $(BDW_GC_LIBS)
libexpr_ORDER_AFTER := $(d)/parser-tab.cc $(d)/parser-tab.hh $(d)/lexer-tab.cc $(d)/lexer-tab.hh
$(d)/parser-tab.cc $(d)/parser-tab.hh: $(d)/parser.y
$(trace-gen) bison -v -o $(libexpr_DIR)/parser-tab.cc $< -d
$(d)/lexer-tab.cc $(d)/lexer-tab.hh: $(d)/lexer.l
$(trace-gen) flex --outfile $(libexpr_DIR)/lexer-tab.cc --header-file=$(libexpr_DIR)/lexer-tab.hh $<
clean-files += $(d)/parser-tab.cc $(d)/parser-tab.hh $(d)/lexer-tab.cc $(d)/lexer-tab.hh
$(eval $(call install-file-in, $(buildprefix)$(d)/nix-expr.pc, $(libdir)/pkgconfig, 0644))
$(foreach i, $(wildcard src/libexpr/value/*.hh), \

View file

@ -1,503 +0,0 @@
%glr-parser
%define api.pure
%locations
%define parse.error verbose
%defines
/* %no-lines */
%parse-param { void * scanner }
%parse-param { nix::ParserState * state }
%lex-param { void * scanner }
%lex-param { nix::ParserState * state }
%expect 1
%expect-rr 1
%code requires {
#ifndef BISON_HEADER
#define BISON_HEADER
#include <variant>
#include "finally.hh"
#include "util.hh"
#include "nixexpr.hh"
#include "eval.hh"
#include "eval-settings.hh"
#include "globals.hh"
#include "parser-state.hh"
#define YYLTYPE ::nix::ParserLocation
#define YY_DECL int yylex \
(YYSTYPE * yylval_param, YYLTYPE * yylloc_param, yyscan_t yyscanner, nix::ParserState * state)
namespace nix {
Expr * parseExprFromBuf(
char * text,
size_t length,
Pos::Origin origin,
const SourcePath & basePath,
SymbolTable & symbols,
PosTable & positions,
const Expr::AstSymbols & astSymbols);
}
#endif
}
%{
#include "parser-tab.hh"
#include "lexer-tab.hh"
YY_DECL;
using namespace nix;
#define CUR_POS state->at(*yylocp)
// otherwise destructors cause compiler errors
#pragma GCC diagnostic ignored "-Wswitch-enum"
#define THROW(err, ...) \
do { \
state->error.reset(new auto(err)); \
[](auto... d) { (delete d, ...); }(__VA_ARGS__); \
YYABORT; \
} while (0)
void yyerror(YYLTYPE * loc, yyscan_t scanner, ParserState * state, const char * error)
{
if (std::string_view(error).starts_with("syntax error, unexpected end of file")) {
loc->first_column = loc->last_column;
loc->first_line = loc->last_line;
}
throw ParseError({
.msg = HintFmt(error),
.pos = state->positions[state->at(*loc)]
});
}
template<typename T>
static std::unique_ptr<T> unp(T * e)
{
return std::unique_ptr<T>(e);
}
template<typename T = std::unique_ptr<nix::Expr>, typename... Args>
static std::vector<T> vec(Args && ... args)
{
std::vector<T> result;
result.reserve(sizeof...(Args));
(result.emplace_back(std::forward<Args>(args)), ...);
return result;
}
%}
%union {
// !!! We're probably leaking stuff here.
nix::Expr * e;
nix::ExprList * list;
nix::ExprAttrs * attrs;
nix::Formals * formals;
nix::Formal * formal;
nix::NixInt n;
nix::NixFloat nf;
nix::StringToken id; // !!! -> Symbol
nix::StringToken path;
nix::StringToken uri;
nix::StringToken str;
std::vector<nix::AttrName> * attrNames;
std::vector<std::pair<nix::AttrName, nix::PosIdx>> * inheritAttrs;
std::vector<std::pair<nix::PosIdx, std::unique_ptr<nix::Expr>>> * string_parts;
std::vector<std::pair<nix::PosIdx, std::variant<std::unique_ptr<nix::Expr>, nix::StringToken>>> * ind_string_parts;
}
%destructor { delete $$; } <e>
%destructor { delete $$; } <list>
%destructor { delete $$; } <attrs>
%destructor { delete $$; } <formals>
%destructor { delete $$; } <formal>
%destructor { delete $$; } <attrNames>
%destructor { delete $$; } <inheritAttrs>
%destructor { delete $$; } <string_parts>
%destructor { delete $$; } <ind_string_parts>
%type <e> start
%type <e> expr expr_function expr_if expr_op
%type <e> expr_select expr_simple expr_app
%type <list> expr_list
%type <attrs> binds
%type <formals> formals
%type <formal> formal
%type <attrNames> attrpath
%type <inheritAttrs> attrs
%type <string_parts> string_parts_interpolated
%type <ind_string_parts> ind_string_parts
%type <e> path_start string_parts string_attr
%type <id> attr
%token <id> ID
%token <str> STR IND_STR
%token <n> INT
%token <nf> FLOAT
%token <path> PATH HPATH SPATH PATH_END
%token <uri> URI
%token IF THEN ELSE ASSERT WITH LET IN REC INHERIT EQ NEQ AND OR IMPL OR_KW
%token DOLLAR_CURLY /* == ${ */
%token IND_STRING_OPEN IND_STRING_CLOSE
%token ELLIPSIS
%right IMPL
%left OR
%left AND
%nonassoc EQ NEQ
%nonassoc '<' '>' LEQ GEQ
%right UPDATE
%left NOT
%left '+' '-'
%left '*' '/'
%right CONCAT
%nonassoc '?'
%nonassoc NEGATE
%%
start: expr { state->result = $1; $$ = 0; };
expr: expr_function;
expr_function
: ID ':' expr_function
{ $$ = new ExprLambda(CUR_POS, state->symbols.create($1), nullptr, unp($3)); }
| '{' formals '}' ':' expr_function
{ if (auto e = state->validateFormals($2)) THROW(*e);
$$ = new ExprLambda(CUR_POS, unp($2), unp($5));
}
| '{' formals '}' '@' ID ':' expr_function
{
auto arg = state->symbols.create($5);
if (auto e = state->validateFormals($2, CUR_POS, arg)) THROW(*e, $2, $7);
$$ = new ExprLambda(CUR_POS, arg, unp($2), unp($7));
}
| ID '@' '{' formals '}' ':' expr_function
{
auto arg = state->symbols.create($1);
if (auto e = state->validateFormals($4, CUR_POS, arg)) THROW(*e, $4, $7);
$$ = new ExprLambda(CUR_POS, arg, unp($4), unp($7));
}
| ASSERT expr ';' expr_function
{ $$ = new ExprAssert(CUR_POS, unp($2), unp($4)); }
| WITH expr ';' expr_function
{ $$ = new ExprWith(CUR_POS, unp($2), unp($4)); }
| LET binds IN expr_function
{ if (!$2->dynamicAttrs.empty())
THROW(ParseError({
.msg = HintFmt("dynamic attributes not allowed in let"),
.pos = state->positions[CUR_POS]
}), $2, $4);
$$ = new ExprLet(unp($2), unp($4));
}
| expr_if
;
expr_if
: IF expr THEN expr ELSE expr { $$ = new ExprIf(CUR_POS, unp($2), unp($4), unp($6)); }
| expr_op
;
expr_op
: '!' expr_op %prec NOT { $$ = new ExprOpNot(unp($2)); }
| '-' expr_op %prec NEGATE { $$ = new ExprCall(CUR_POS, std::make_unique<ExprVar>(state->s.sub), vec(std::make_unique<ExprInt>(0), unp($2))); }
| expr_op EQ expr_op { $$ = new ExprOpEq(unp($1), unp($3)); }
| expr_op NEQ expr_op { $$ = new ExprOpNEq(unp($1), unp($3)); }
| expr_op '<' expr_op { $$ = new ExprCall(state->at(@2), std::make_unique<ExprVar>(state->s.lessThan), vec($1, $3)); }
| expr_op LEQ expr_op { $$ = new ExprOpNot(std::make_unique<ExprCall>(state->at(@2), std::make_unique<ExprVar>(state->s.lessThan), vec($3, $1))); }
| expr_op '>' expr_op { $$ = new ExprCall(state->at(@2), std::make_unique<ExprVar>(state->s.lessThan), vec($3, $1)); }
| expr_op GEQ expr_op { $$ = new ExprOpNot(std::make_unique<ExprCall>(state->at(@2), std::make_unique<ExprVar>(state->s.lessThan), vec($1, $3))); }
| expr_op AND expr_op { $$ = new ExprOpAnd(state->at(@2), unp($1), unp($3)); }
| expr_op OR expr_op { $$ = new ExprOpOr(state->at(@2), unp($1), unp($3)); }
| expr_op IMPL expr_op { $$ = new ExprOpImpl(state->at(@2), unp($1), unp($3)); }
| expr_op UPDATE expr_op { $$ = new ExprOpUpdate(state->at(@2), unp($1), unp($3)); }
| expr_op '?' attrpath { $$ = new ExprOpHasAttr(unp($1), std::move(*$3)); delete $3; }
| expr_op '+' expr_op
{ $$ = new ExprConcatStrings(state->at(@2), false, vec<std::pair<PosIdx, std::unique_ptr<Expr>>>(std::pair(state->at(@1), unp($1)), std::pair(state->at(@3), unp($3)))); }
| expr_op '-' expr_op { $$ = new ExprCall(state->at(@2), std::make_unique<ExprVar>(state->s.sub), vec($1, $3)); }
| expr_op '*' expr_op { $$ = new ExprCall(state->at(@2), std::make_unique<ExprVar>(state->s.mul), vec($1, $3)); }
| expr_op '/' expr_op { $$ = new ExprCall(state->at(@2), std::make_unique<ExprVar>(state->s.div), vec($1, $3)); }
| expr_op CONCAT expr_op { $$ = new ExprOpConcatLists(state->at(@2), unp($1), unp($3)); }
| expr_app
;
expr_app
: expr_app expr_select {
if (auto e2 = dynamic_cast<ExprCall *>($1)) {
e2->args.emplace_back($2);
$$ = $1;
} else
$$ = new ExprCall(CUR_POS, unp($1), vec(unp($2)));
}
| expr_select
;
expr_select
: expr_simple '.' attrpath
{ $$ = new ExprSelect(CUR_POS, unp($1), std::move(*$3), nullptr); delete $3; }
| expr_simple '.' attrpath OR_KW expr_select
{ $$ = new ExprSelect(CUR_POS, unp($1), std::move(*$3), unp($5)); delete $3; }
| /* Backwards compatibility: because Nixpkgs has a rarely used
function named or, allow stuff like map or [...]. */
expr_simple OR_KW
{ $$ = new ExprCall(CUR_POS, unp($1), vec(std::make_unique<ExprVar>(CUR_POS, state->s.or_))); }
| expr_simple
;
expr_simple
: ID {
std::string_view s = "__curPos";
if ($1.l == s.size() && strncmp($1.p, s.data(), s.size()) == 0)
$$ = new ExprPos(CUR_POS);
else
$$ = new ExprVar(CUR_POS, state->symbols.create($1));
}
| INT { $$ = new ExprInt($1); }
| FLOAT { $$ = new ExprFloat($1); }
| '"' string_parts '"' { $$ = $2; }
| IND_STRING_OPEN ind_string_parts IND_STRING_CLOSE {
$$ = state->stripIndentation(CUR_POS, std::move(*$2)).release();
delete $2;
}
| path_start PATH_END
| path_start string_parts_interpolated PATH_END {
$2->emplace($2->begin(), state->at(@1), $1);
$$ = new ExprConcatStrings(CUR_POS, false, std::move(*$2));
delete $2;
}
| SPATH {
std::string path($1.p + 1, $1.l - 2);
$$ = new ExprCall(CUR_POS,
std::make_unique<ExprVar>(state->s.findFile),
vec(std::make_unique<ExprVar>(state->s.nixPath),
std::make_unique<ExprString>(std::move(path))));
}
| URI {
static bool noURLLiterals = experimentalFeatureSettings.isEnabled(Xp::NoUrlLiterals);
if (noURLLiterals)
THROW(ParseError({
.msg = HintFmt("URL literals are disabled"),
.pos = state->positions[CUR_POS]
}));
$$ = new ExprString(std::string($1));
}
| '(' expr ')' { $$ = $2; }
/* Let expressions `let {..., body = ...}' are just desugared
into `(rec {..., body = ...}).body'. */
| LET '{' binds '}'
{ $3->recursive = true; $$ = new ExprSelect(noPos, unp($3), state->s.body); }
| REC '{' binds '}'
{ $3->recursive = true; $$ = $3; }
| '{' binds '}'
{ $$ = $2; }
| '[' expr_list ']' { $$ = $2; }
;
string_parts
: STR { $$ = new ExprString(std::string($1)); }
| string_parts_interpolated
{ $$ = new ExprConcatStrings(CUR_POS, true, std::move(*$1));
delete $1;
}
| { $$ = new ExprString(""); }
;
string_parts_interpolated
: string_parts_interpolated STR
{ $$ = $1; $1->emplace_back(state->at(@2), new ExprString(std::string($2))); }
| string_parts_interpolated DOLLAR_CURLY expr '}' { $$ = $1; $1->emplace_back(state->at(@2), $3); }
| DOLLAR_CURLY expr '}' { $$ = new std::vector<std::pair<PosIdx, std::unique_ptr<Expr>>>; $$->emplace_back(state->at(@1), $2); }
| STR DOLLAR_CURLY expr '}' {
$$ = new std::vector<std::pair<PosIdx, std::unique_ptr<Expr>>>;
$$->emplace_back(state->at(@1), new ExprString(std::string($1)));
$$->emplace_back(state->at(@2), $3);
}
;
path_start
: PATH {
Path path(absPath({$1.p, $1.l}, state->basePath.path.abs()));
/* add back in the trailing '/' to the first segment */
if ($1.p[$1.l-1] == '/' && $1.l > 1)
path += "/";
$$ = new ExprPath(path);
}
| HPATH {
if (evalSettings.pureEval) {
THROW(Error(
"the path '%s' can not be resolved in pure mode",
std::string_view($1.p, $1.l)
));
}
Path path(getHome() + std::string($1.p + 1, $1.l - 1));
$$ = new ExprPath(path);
}
;
ind_string_parts
: ind_string_parts IND_STR { $$ = $1; $1->emplace_back(state->at(@2), $2); }
| ind_string_parts DOLLAR_CURLY expr '}' { $$ = $1; $1->emplace_back(state->at(@2), unp($3)); }
| { $$ = new std::vector<std::pair<PosIdx, std::variant<std::unique_ptr<Expr>, StringToken>>>; }
;
binds
: binds attrpath '=' expr ';'
{ $$ = $1;
if (auto e = state->addAttr($$, std::move(*$2), unp($4), state->at(@2))) THROW(*e, $1, $2);
delete $2;
}
| binds INHERIT attrs ';'
{ $$ = $1;
for (auto & [i, iPos] : *$3) {
if ($$->attrs.find(i.symbol) != $$->attrs.end())
THROW(state->dupAttr(i.symbol, iPos, $$->attrs[i.symbol].pos), $1);
$$->attrs.emplace(
i.symbol,
ExprAttrs::AttrDef(std::make_unique<ExprVar>(iPos, i.symbol), iPos, ExprAttrs::AttrDef::Kind::Inherited));
}
delete $3;
}
| binds INHERIT '(' expr ')' attrs ';'
{ $$ = $1;
if (!$$->inheritFromExprs)
$$->inheritFromExprs = std::make_unique<std::vector<std::unique_ptr<Expr>>>();
$$->inheritFromExprs->push_back(unp($4));
for (auto & [i, iPos] : *$6) {
if ($$->attrs.find(i.symbol) != $$->attrs.end())
THROW(state->dupAttr(i.symbol, iPos, $$->attrs[i.symbol].pos), $1);
auto from = std::make_unique<nix::ExprInheritFrom>(state->at(@4), $$->inheritFromExprs->size() - 1);
$$->attrs.emplace(
i.symbol,
ExprAttrs::AttrDef(
std::make_unique<ExprSelect>(iPos, std::move(from), i.symbol),
iPos,
ExprAttrs::AttrDef::Kind::InheritedFrom));
}
delete $6;
}
| { $$ = new ExprAttrs(state->at(@0)); }
;
attrs
: attrs attr { $$ = $1; $1->emplace_back(AttrName(state->symbols.create($2)), state->at(@2)); }
| attrs string_attr
{ $$ = $1;
ExprString * str = dynamic_cast<ExprString *>($2);
if (str) {
$$->emplace_back(AttrName(state->symbols.create(str->s)), state->at(@2));
delete str;
} else
THROW(ParseError({
.msg = HintFmt("dynamic attributes not allowed in inherit"),
.pos = state->positions[state->at(@2)]
}), $1, $2);
}
| { $$ = new std::vector<std::pair<AttrName, PosIdx>>; }
;
attrpath
: attrpath '.' attr { $$ = $1; $1->push_back(AttrName(state->symbols.create($3))); }
| attrpath '.' string_attr
{ $$ = $1;
ExprString * str = dynamic_cast<ExprString *>($3);
if (str) {
$$->push_back(AttrName(state->symbols.create(str->s)));
delete str;
} else
$$->emplace_back(unp($3));
}
| attr { $$ = new std::vector<AttrName>; $$->push_back(AttrName(state->symbols.create($1))); }
| string_attr
{ $$ = new std::vector<AttrName>;
ExprString *str = dynamic_cast<ExprString *>($1);
if (str) {
$$->push_back(AttrName(state->symbols.create(str->s)));
delete str;
} else
$$->emplace_back(unp($1));
}
;
attr
: ID
| OR_KW { $$ = {"or", 2}; }
;
string_attr
: '"' string_parts '"' { $$ = $2; }
| DOLLAR_CURLY expr '}' { $$ = $2; }
;
expr_list
: expr_list expr_select { $$ = $1; $1->elems.emplace_back($2); /* !!! dangerous */ }
| { $$ = new ExprList; }
;
formals
: formal ',' formals
{ $$ = $3; $$->formals.emplace_back(std::move(*$1)); delete $1; }
| formal
{ $$ = new Formals; $$->formals.emplace_back(std::move(*$1)); $$->ellipsis = false; delete $1; }
|
{ $$ = new Formals; $$->ellipsis = false; }
| ELLIPSIS
{ $$ = new Formals; $$->ellipsis = true; }
;
formal
: ID { $$ = new Formal{CUR_POS, state->symbols.create($1), nullptr}; }
| ID '?' expr { $$ = new Formal{CUR_POS, state->symbols.create($1), unp($3)}; }
;
%%
#include "eval.hh"
namespace nix {
Expr * parseExprFromBuf(
char * text,
size_t length,
Pos::Origin origin,
const SourcePath & basePath,
SymbolTable & symbols,
PosTable & positions,
const Expr::AstSymbols & astSymbols)
{
yyscan_t scanner;
ParserState state {
.symbols = symbols,
.positions = positions,
.basePath = basePath,
.origin = positions.addOrigin(origin, length),
.s = astSymbols,
};
yylex_init(&scanner);
Finally _destroy([&] { yylex_destroy(scanner); });
yy_scan_buffer(text, length, scanner);
yyparse(scanner, &state);
if (state.error) {
delete state.result;
throw *state.error;
}
return state.result;
}
}

View file

@ -0,0 +1,65 @@
#pragma once
#include <tao/pegtl.hpp>
namespace nix::parser {
// modified copy of change_state, as the manual suggest for more involved
// state manipulation. we want to change only the first state parameter,
// and we care about the *initial* position of a rule application (not the
// past-the-end position as pegtl change_state provides)
template<typename NewState>
struct change_head : tao::pegtl::maybe_nothing
{
template<
typename Rule,
tao::pegtl::apply_mode A,
tao::pegtl::rewind_mode M,
template<typename...> class Action,
template<typename...> class Control,
typename ParseInput,
typename State,
typename... States
>
[[nodiscard]] static bool match(ParseInput & in, State && st, States &&... sts)
{
const auto begin = in.iterator();
if constexpr (std::is_constructible_v<NewState, State, States...>) {
NewState s(st, sts...);
if (tao::pegtl::match<Rule, A, M, Action, Control>(in, s, sts...)) {
if constexpr (A == tao::pegtl::apply_mode::action) {
_success<Action<Rule>>(0, begin, in, s, st, sts...);
}
return true;
}
return false;
} else if constexpr (std::is_default_constructible_v<NewState>) {
NewState s;
if (tao::pegtl::match<Rule, A, M, Action, Control>(in, s, sts...)) {
if constexpr (A == tao::pegtl::apply_mode::action) {
_success<Action<Rule>>(0, begin, in, s, st, sts...);
}
return true;
}
return false;
} else {
static_assert(decltype(sizeof(NewState))(), "unable to instantiate new state");
}
}
template<typename Target, typename ParseInput, typename... S>
static void _success(void *, auto & begin, ParseInput & in, S & ... sts)
{
const typename ParseInput::action_t at(begin, in);
Target::success(at, sts...);
}
template<typename Target, typename... S>
static void _success(decltype(Target::success0(std::declval<S &>()...), 0), auto &, auto &, S & ... sts)
{
Target::success0(sts...);
}
};
}

View file

@ -0,0 +1,706 @@
#pragma once
#include "tao/pegtl.hpp"
#include <type_traits>
#include <variant>
#include <boost/container/small_vector.hpp>
// NOTE
// nix line endings are \n, \r\n, \r. the grammar does not use eol or
// eolf rules in favor of reproducing the old flex lexer as faithfully as
// possible, and deferring calculation of positions to downstream users.
namespace nix::parser::grammar {
using namespace tao::pegtl;
namespace p = tao::pegtl;
// character classes
namespace c {
struct path : sor<
ranges<'a', 'z', 'A', 'Z', '0', '9'>,
one<'.', '_', '-', '+'>
> {};
struct path_sep : one<'/'> {};
struct id_first : ranges<'a', 'z', 'A', 'Z', '_'> {};
struct id_rest : sor<
ranges<'a', 'z', 'A', 'Z', '0', '9'>,
one<'_', '\'', '-'>
> {};
struct uri_scheme_first : ranges<'a', 'z', 'A', 'Z'> {};
struct uri_scheme_rest : sor<
ranges<'a', 'z', 'A', 'Z', '0', '9'>,
one<'+', '-', '.'>
> {};
struct uri_sep : one<':'> {};
struct uri_rest : sor<
ranges<'a', 'z', 'A', 'Z', '0', '9'>,
one<'%', '/', '?', ':', '@', '&', '=', '+', '$', ',', '-', '_', '.', '!', '~', '*', '\''>
> {};
}
// "tokens". PEGs don't really care about tokens, we merely use them as a convenient
// way of writing down keywords and a couple complicated syntax rules.
namespace t {
struct _extend_as_path : seq<
star<c::path>,
not_at<TAO_PEGTL_STRING("/*")>,
not_at<TAO_PEGTL_STRING("//")>,
c::path_sep,
sor<c::path, TAO_PEGTL_STRING("${")>
> {};
struct _extend_as_uri : seq<
star<c::uri_scheme_rest>,
c::uri_sep,
c::uri_rest
> {};
// keywords might be extended to identifiers, paths, or uris.
// NOTE this assumes that keywords are a-zA-Z only, otherwise uri schemes would never
// match correctly.
// NOTE not a simple seq<...> because this would report incorrect positions for
// keywords used inside must<> if a prefix of the keyword matches.
template<typename S>
struct _keyword : sor<
seq<
S,
not_at<c::id_rest>,
not_at<_extend_as_path>,
not_at<_extend_as_uri>
>,
failure
> {};
struct kw_if : _keyword<TAO_PEGTL_STRING("if")> {};
struct kw_then : _keyword<TAO_PEGTL_STRING("then")> {};
struct kw_else : _keyword<TAO_PEGTL_STRING("else")> {};
struct kw_assert : _keyword<TAO_PEGTL_STRING("assert")> {};
struct kw_with : _keyword<TAO_PEGTL_STRING("with")> {};
struct kw_let : _keyword<TAO_PEGTL_STRING("let")> {};
struct kw_in : _keyword<TAO_PEGTL_STRING("in")> {};
struct kw_rec : _keyword<TAO_PEGTL_STRING("rec")> {};
struct kw_inherit : _keyword<TAO_PEGTL_STRING("inherit")> {};
struct kw_or : _keyword<TAO_PEGTL_STRING("or")> {};
// `-` can be a unary prefix op, a binary infix op, or the first character
// of a path or -> (ex 1->1--1)
// `/` can be a path leader or an operator (ex a?a /a)
struct op_minus : seq<one<'-'>, not_at<one<'>'>>, not_at<_extend_as_path>> {};
struct op_div : seq<one<'/'>, not_at<c::path>> {};
// match a rule, making sure we are not matching it where a keyword would match.
// using minus like this is a lot faster than flipping the order and using seq.
template<typename... Rules>
struct _not_at_any_keyword : minus<
seq<Rules...>,
sor<
TAO_PEGTL_STRING("inherit"),
TAO_PEGTL_STRING("assert"),
TAO_PEGTL_STRING("else"),
TAO_PEGTL_STRING("then"),
TAO_PEGTL_STRING("with"),
TAO_PEGTL_STRING("let"),
TAO_PEGTL_STRING("rec"),
TAO_PEGTL_STRING("if"),
TAO_PEGTL_STRING("in"),
TAO_PEGTL_STRING("or")
>
> {};
// identifiers are kind of horrid:
//
// - uri_scheme_first ⊂ id_first
// - uri_scheme_first ⊂ uri_scheme_rest ⊂ path
// - id_first ⊂ id_rest { ' } ⊂ path
// - id_first ∩ (path uri_scheme_first) = { _ }
// - uri_sep ∉ { id_first, id_rest, uri_scheme_first, uri_scheme_rest, path }
// - path_sep ∉ { id_first, id_rest, uri_scheme_first, uri_scheme_rest }
//
// and we want, without reading the input more than once, a string that
// matches (id_first id_rest*) and is not followed by any number of
// characters such that the extended string matches path or uri rules.
//
// since the first character must be either _ or a uri scheme character
// we can ignore path-like bits at the beginning. uri_sep cannot appear anywhere
// in an identifier, so it's only needed in lookahead checks at the uri-like
// prefix. likewise path_sep cannot appear anywhere in the idenfier, so it's
// only needed in lookahead checks in the path-like prefix.
//
// in total that gives us a decomposition of
//
// (uri-scheme-like? (?! continues-as-uri) | _)
// (path-segment-like? (?! continues-as-path))
// id_rest*
struct identifier : _not_at_any_keyword<
// we don't use (at<id_rest>, ...) matches here because identifiers are
// a really hot path and rewinding as needed by at<> isn't entirely free.
sor<
seq<
c::uri_scheme_first,
star<ranges<'a', 'z', 'A', 'Z', '0', '9', '-'>>,
not_at<_extend_as_uri>
>,
one<'_'>
>,
star<sor<ranges<'a', 'z', 'A', 'Z', '0', '9'>, one<'_', '-'>>>,
not_at<_extend_as_path>,
star<c::id_rest>
> {};
// floats may extend ints, thus these rules are very similar.
struct integer : seq<
sor<
seq<range<'1', '9'>, star<digit>, not_at<one<'.'>>>,
seq<one<'0'>, not_at<one<'.'>, digit>, star<digit>>
>,
not_at<_extend_as_path>
> {};
struct floating : seq<
sor<
seq<range<'1', '9'>, star<digit>, one<'.'>, star<digit>>,
seq<opt<one<'0'>>, one<'.'>, plus<digit>>
>,
opt<one<'E', 'e'>, opt<one<'+', '-'>>, plus<digit>>,
not_at<_extend_as_path>
> {};
struct uri : seq<
c::uri_scheme_first,
star<c::uri_scheme_rest>,
c::uri_sep,
plus<c::uri_rest>
> {};
struct sep : sor<
plus<one<' ', '\t', '\r', '\n'>>,
seq<one<'#'>, star<not_one<'\r', '\n'>>>,
seq<string<'/', '*'>, until<string<'*', '/'>>>
> {};
}
using seps = star<t::sep>;
// marker for semantic rules. not handling one of these in an action that cares about
// semantics is probably an error.
struct semantic {};
struct expr;
struct _string {
template<typename... Inner>
struct literal : semantic, seq<Inner...> {};
struct cr_lf : semantic, seq<one<'\r'>, opt<one<'\n'>>> {};
struct interpolation : semantic, seq<
p::string<'$', '{'>, seps,
must<expr>, seps,
must<one<'}'>>
> {};
struct escape : semantic, must<any> {};
};
struct string : _string, seq<
one<'"'>,
star<
sor<
_string::literal<plus<not_one<'$', '"', '\\', '\r'>>>,
_string::cr_lf,
_string::interpolation,
_string::literal<one<'$'>, opt<one<'$'>>>,
seq<one<'\\'>, _string::escape>
>
>,
must<one<'"'>>
> {};
struct _ind_string {
template<bool Indented, typename... Inner>
struct literal : semantic, seq<Inner...> {};
struct interpolation : semantic, seq<
p::string<'$', '{'>, seps,
must<expr>, seps,
must<one<'}'>>
> {};
struct escape : semantic, must<any> {};
};
struct ind_string : _ind_string, seq<
TAO_PEGTL_STRING("''"),
opt<star<one<' '>>, one<'\n'>>,
star<
sor<
_ind_string::literal<
true,
plus<
sor<
not_one<'$', '\''>,
seq<one<'$'>, not_one<'{', '\''>>,
seq<one<'\''>, not_one<'\'', '$'>>
>
>
>,
_ind_string::interpolation,
_ind_string::literal<false, one<'$'>>,
_ind_string::literal<false, one<'\''>, not_at<one<'\''>>>,
seq<one<'\''>, _ind_string::literal<false, p::string<'\'', '\''>>>,
seq<
p::string<'\'', '\''>,
sor<
_ind_string::literal<false, one<'$'>>,
seq<one<'\\'>, _ind_string::escape>
>
>
>
>,
must<TAO_PEGTL_STRING("''")>
> {};
struct _path {
// legacy lexer rules. extra l_ to avoid reserved c++ identifiers.
struct _l_PATH : seq<star<c::path>, plus<c::path_sep, plus<c::path>>, opt<c::path_sep>> {};
struct _l_PATH_SEG : seq<star<c::path>, c::path_sep> {};
struct _l_HPATH : seq<one<'~'>, plus<c::path_sep, plus<c::path>>, opt<c::path_sep>> {};
struct _l_HPATH_START : TAO_PEGTL_STRING("~/") {};
struct _path_str : sor<_l_PATH, _l_PATH_SEG, plus<c::path>> {};
// modern rules
template<typename... Inner>
struct literal : semantic, seq<Inner...> {};
struct interpolation : semantic, seq<
p::string<'$', '{'>, seps,
must<expr>, seps,
must<one<'}'>>
> {};
struct anchor : semantic, sor<
_l_PATH,
seq<_l_PATH_SEG, at<TAO_PEGTL_STRING("${")>>
> {};
struct home_anchor : semantic, sor<
_l_HPATH,
seq<_l_HPATH_START, at<TAO_PEGTL_STRING("${")>>
> {};
struct searched_path : semantic, list<plus<c::path>, c::path_sep> {};
struct forbid_prefix_triple_slash : sor<not_at<c::path_sep>, failure> {};
struct forbid_prefix_double_slash_no_interp : sor<
not_at<c::path_sep, star<c::path>, not_at<TAO_PEGTL_STRING("${")>>,
failure
> {};
// legacy parser rules
struct _str_rest : seq<
must<forbid_prefix_double_slash_no_interp>,
opt<literal<_path_str>>,
must<forbid_prefix_triple_slash>,
star<
sor<
literal<_path_str>,
interpolation
>
>
> {};
};
struct path : _path, sor<
seq<
sor<_path::anchor, _path::home_anchor>,
_path::_str_rest
>,
seq<one<'<'>, _path::searched_path, one<'>'>>
> {};
struct _formal {
struct name : semantic, t::identifier {};
struct default_value : semantic, must<expr> {};
};
struct formal : semantic, _formal, seq<
_formal::name,
opt<seps, one<'?'>, seps, _formal::default_value>
> {};
struct _formals {
struct ellipsis : semantic, p::ellipsis {};
};
struct formals : semantic, _formals, seq<
one<'{'>, seps,
// formals and attrsets share a two-token head sequence ('{' <id>).
// this rule unrolls the formals list a bit to provide better error messages than
// "expected '='" at the first ',' if formals are incorrect.
sor<
one<'}'>,
seq<_formals::ellipsis, seps, must<one<'}'>>>,
seq<
formal, seps,
if_then_else<
at<one<','>>,
seq<
star<one<','>, seps, formal, seps>,
opt<one<','>, seps, opt<_formals::ellipsis, seps>>,
must<one<'}'>>
>,
one<'}'>
>
>
>
> {};
struct _attr {
struct simple : semantic, sor<t::identifier, t::kw_or> {};
struct string : semantic, seq<grammar::string> {};
struct expr : semantic, seq<
TAO_PEGTL_STRING("${"), seps,
must<grammar::expr>, seps,
must<one<'}'>>
> {};
};
struct attr : _attr, sor<
_attr::simple,
_attr::string,
_attr::expr
> {};
struct attrpath : list<attr, one<'.'>, t::sep> {};
struct _inherit {
struct from : semantic, must<expr> {};
struct attrs : list<attr, seps> {};
};
struct inherit : _inherit, seq<
t::kw_inherit, seps,
opt<one<'('>, seps, _inherit::from, seps, must<one<')'>>, seps>,
opt<_inherit::attrs, seps>,
must<one<';'>>
> {};
struct _binding {
struct path : semantic, attrpath {};
struct equal : one<'='> {};
struct value : semantic, must<expr> {};
};
struct binding : _binding, seq<
_binding::path, seps,
must<_binding::equal>, seps,
_binding::value, seps,
must<one<';'>>
> {};
struct bindings : opt<list<sor<inherit, binding>, seps>> {};
struct op {
enum class kind {
// NOTE non-associativity is *NOT* handled in the grammar structure.
// handling it in the grammar itself instead of in semantic actions
// slows down the parser significantly and makes the rules *much*
// harder to read. maybe this will be different at some point when
// ! does not sit between two binary precedence levels.
nonAssoc,
leftAssoc,
rightAssoc,
unary,
};
template<typename Rule, unsigned Precedence, kind Kind = kind::leftAssoc>
struct _op : Rule {
static constexpr unsigned precedence = Precedence;
static constexpr op::kind kind = Kind;
};
struct unary_minus : _op<t::op_minus, 3, kind::unary> {};
// treating this like a unary postfix operator is sketchy, but that's
// the most reasonable way to implement the operator precedence set forth
// by the language way back. it'd be much better if `.` and `?` had the same
// precedence, but alas.
struct has_attr : _op<seq<one<'?'>, seps, must<attrpath>>, 4> {};
struct concat : _op<TAO_PEGTL_STRING("++"), 5, kind::rightAssoc> {};
struct mul : _op<one<'*'>, 6> {};
struct div : _op<t::op_div, 6> {};
struct plus : _op<one<'+'>, 7> {};
struct minus : _op<t::op_minus, 7> {};
struct not_ : _op<one<'!'>, 8, kind::unary> {};
struct update : _op<TAO_PEGTL_STRING("//"), 9, kind::rightAssoc> {};
struct less_eq : _op<TAO_PEGTL_STRING("<="), 10, kind::nonAssoc> {};
struct greater_eq : _op<TAO_PEGTL_STRING(">="), 10, kind::nonAssoc> {};
struct less : _op<one<'<'>, 10, kind::nonAssoc> {};
struct greater : _op<one<'>'>, 10, kind::nonAssoc> {};
struct equals : _op<TAO_PEGTL_STRING("=="), 11, kind::nonAssoc> {};
struct not_equals : _op<TAO_PEGTL_STRING("!="), 11, kind::nonAssoc> {};
struct and_ : _op<TAO_PEGTL_STRING("&&"), 12> {};
struct or_ : _op<TAO_PEGTL_STRING("||"), 13> {};
struct implies : _op<TAO_PEGTL_STRING("->"), 14, kind::rightAssoc> {};
};
struct _expr {
template<template<typename...> class OpenMod = seq, typename... Init>
struct _attrset : seq<
Init...,
OpenMod<one<'{'>>, seps,
bindings, seps,
must<one<'}'>>
> {};
struct select;
struct id : semantic, t::identifier {};
struct int_ : semantic, t::integer {};
struct float_ : semantic, t::floating {};
struct string : semantic, seq<grammar::string> {};
struct ind_string : semantic, seq<grammar::ind_string> {};
struct path : semantic, seq<grammar::path> {};
struct uri : semantic, t::uri {};
struct ancient_let : semantic, _attrset<must, t::kw_let, seps> {};
struct rec_set : semantic, _attrset<must, t::kw_rec, seps> {};
struct set : semantic, _attrset<> {};
struct _list {
struct entry : semantic, seq<select> {};
};
struct list : semantic, _list, seq<
one<'['>, seps,
opt<p::list<_list::entry, seps>, seps>,
must<one<']'>>
> {};
struct _simple : sor<
id,
int_,
float_,
string,
ind_string,
path,
uri,
seq<one<'('>, seps, must<expr>, seps, must<one<')'>>>,
ancient_let,
rec_set,
set,
list
> {};
struct _select {
struct head : _simple {};
struct attr : semantic, seq<attrpath> {};
struct attr_or : semantic, must<select> {};
struct as_app_or : semantic, t::kw_or {};
};
struct _app {
struct first_arg : semantic, seq<select> {};
struct another_arg : semantic, seq<select> {};
// can be used to stash a position of the application head node
struct select_or_fn : seq<select> {};
};
struct select : _select, seq<
_select::head, seps,
opt<
sor<
seq<
one<'.'>, seps, _select::attr,
opt<seps, t::kw_or, seps, _select::attr_or>
>,
_select::as_app_or
>
>
> {};
struct app : _app, seq<
_app::select_or_fn,
opt<seps, _app::first_arg, star<seps, _app::another_arg>>
> {};
template<typename Op>
struct operator_ : semantic, Op {};
struct unary : seq<
star<sor<operator_<op::not_>, operator_<op::unary_minus>>, seps>,
app
> {};
struct _binary_operator : sor<
operator_<op::implies>,
operator_<op::update>,
operator_<op::concat>,
operator_<op::plus>,
operator_<op::minus>,
operator_<op::mul>,
operator_<op::div>,
operator_<op::less_eq>,
operator_<op::greater_eq>,
operator_<op::less>,
operator_<op::greater>,
operator_<op::equals>,
operator_<op::not_equals>,
operator_<op::or_>,
operator_<op::and_>
> {};
struct _binop : seq<
unary,
star<
seps,
sor<
seq<_binary_operator, seps, must<unary>>,
operator_<op::has_attr>
>
>
> {};
struct _lambda {
struct arg : semantic, t::identifier {};
};
struct lambda : semantic, _lambda, sor<
seq<
_lambda::arg, seps,
sor<
seq<one<':'>, seps, must<expr>>,
seq<one<'@'>, seps, must<formals, seps, one<':'>, seps, expr>>
>
>,
seq<
formals, seps,
sor<
seq<one<':'>, seps, must<expr>>,
seq<one<'@'>, seps, must<_lambda::arg, seps, one<':'>, seps, expr>>
>
>
> {};
struct assert_ : semantic, seq<
t::kw_assert, seps,
must<expr>, seps,
must<one<';'>>, seps,
must<expr>
> {};
struct with : semantic, seq<
t::kw_with, seps,
must<expr>, seps,
must<one<';'>>, seps,
must<expr>
> {};
struct let : seq<
t::kw_let, seps,
not_at<one<'{'>>, // exclude ancient_let so we can must<kw_in>
bindings, seps,
must<t::kw_in>, seps,
must<expr>
> {};
struct if_ : semantic, seq<
t::kw_if, seps,
must<expr>, seps,
must<t::kw_then>, seps,
must<expr>, seps,
must<t::kw_else>, seps,
must<expr>
> {};
};
struct expr : semantic, _expr, sor<
_expr::lambda,
_expr::assert_,
_expr::with,
_expr::let,
_expr::if_,
_expr::_binop
> {};
// legacy support: \0 terminates input if passed from flex to bison as a token
struct eof : sor<p::eof, one<0>> {};
struct root : must<seps, expr, seps, eof> {};
template<typename Rule>
struct nothing : p::nothing<Rule> {
static_assert(!std::is_base_of_v<semantic, Rule>);
};
template<typename Self, typename OpCtx, typename AttrPathT, typename ExprT>
struct operator_semantics {
struct has_attr : grammar::op::has_attr {
AttrPathT path;
};
struct OpEntry {
OpCtx ctx;
uint8_t prec;
grammar::op::kind assoc;
std::variant<
grammar::op::not_,
grammar::op::unary_minus,
grammar::op::implies,
grammar::op::or_,
grammar::op::and_,
grammar::op::equals,
grammar::op::not_equals,
grammar::op::less_eq,
grammar::op::greater_eq,
grammar::op::update,
grammar::op::concat,
grammar::op::less,
grammar::op::greater,
grammar::op::plus,
grammar::op::minus,
grammar::op::mul,
grammar::op::div,
has_attr
> op;
};
// statistics here are taken from nixpkgs commit de502c4d0ba96261e5de803e4d1d1925afd3e22f.
// over 99.9% of contexts in nixpkgs need at most 4 slots, ~85% need only 1
boost::container::small_vector<ExprT, 4> exprs;
// over 99.9% of contexts in nixpkgs need at most 2 slots, ~85% need only 1
boost::container::small_vector<OpEntry, 2> ops;
// derived class is expected to define members:
//
// ExprT applyOp(OpCtx & pos, auto & op, auto &... args);
// [[noreturn]] static void badOperator(OpCtx & pos, auto &... args);
void reduce(uint8_t toPrecedence, auto &... args) {
while (!ops.empty()) {
auto & [ctx, precedence, kind, op] = ops.back();
// NOTE this relies on associativity not being mixed within a precedence level.
if ((precedence > toPrecedence)
|| (kind != grammar::op::kind::leftAssoc && precedence == toPrecedence))
break;
std::visit([&, ctx=std::move(ctx)] (auto & op) {
exprs.push_back(static_cast<Self &>(*this).applyOp(ctx, op, args...));
}, op);
ops.pop_back();
}
}
ExprT popExpr()
{
auto r = std::move(exprs.back());
exprs.pop_back();
return r;
}
void pushOp(OpCtx ctx, auto o, auto &... args)
{
if (o.kind != grammar::op::kind::unary)
reduce(o.precedence, args...);
if (!ops.empty() && o.kind == grammar::op::kind::nonAssoc) {
auto & [_pos, _prec, _kind, _o] = ops.back();
if (_kind == o.kind && _prec == o.precedence)
Self::badOperator(ctx, args...);
}
ops.emplace_back(ctx, o.precedence, o.kind, std::move(o));
}
ExprT finish(auto &... args)
{
reduce(255, args...);
return popExpr();
}
};
}

View file

@ -0,0 +1,852 @@
#include "attr-set.hh"
#include "error.hh"
#include "eval.hh"
#include "eval-settings.hh"
#include "finally.hh"
#include "nixexpr.hh"
#include "symbol-table.hh"
#include "change_head.hh"
#include "grammar.hh"
#include "state.hh"
#include <charconv>
#include <clocale>
#include <memory>
// flip this define when doing parser development to enable some g checks.
#if 0
#include <tao/pegtl/contrib/analyze.hpp>
#define ANALYZE_GRAMMAR \
([] { \
const std::size_t issues = tao::pegtl::analyze<grammar::root>(); \
assert(issues == 0); \
})()
#else
#define ANALYZE_GRAMMAR ((void) 0)
#endif
namespace p = tao::pegtl;
namespace nix::parser {
namespace {
template<typename>
inline constexpr const char * error_message = nullptr;
#define error_message_for(...) \
template<> inline constexpr auto error_message<__VA_ARGS__>
error_message_for(p::one<'{'>) = "expecting '{'";
error_message_for(p::one<'}'>) = "expecting '}'";
error_message_for(p::one<'"'>) = "expecting '\"'";
error_message_for(p::one<';'>) = "expecting ';'";
error_message_for(p::one<')'>) = "expecting ')'";
error_message_for(p::one<'='>) = "expecting '='";
error_message_for(p::one<']'>) = "expecting ']'";
error_message_for(p::one<':'>) = "expecting ':'";
error_message_for(p::string<'\'', '\''>) = "expecting \"''\"";
error_message_for(p::any) = "expecting any character";
error_message_for(grammar::eof) = "expecting end of file";
error_message_for(grammar::seps) = "expecting separators";
error_message_for(grammar::path::forbid_prefix_triple_slash) = "too many slashes in path";
error_message_for(grammar::path::forbid_prefix_double_slash_no_interp) = "path has a trailing slash";
error_message_for(grammar::expr) = "expecting expression";
error_message_for(grammar::expr::unary) = "expecting expression";
error_message_for(grammar::binding::equal) = "expecting '='";
error_message_for(grammar::expr::lambda::arg) = "expecting identifier";
error_message_for(grammar::formals) = "expecting formals";
error_message_for(grammar::attrpath) = "expecting attribute path";
error_message_for(grammar::expr::select) = "expecting selection expression";
error_message_for(grammar::t::kw_then) = "expecting 'then'";
error_message_for(grammar::t::kw_else) = "expecting 'else'";
error_message_for(grammar::t::kw_in) = "expecting 'in'";
struct SyntaxErrors
{
template<typename Rule>
static constexpr auto message = error_message<Rule>;
template<typename Rule>
static constexpr bool raise_on_failure = false;
};
template<typename Rule>
struct Control : p::must_if<SyntaxErrors>::control<Rule>
{
template<typename ParseInput, typename... States>
[[noreturn]] static void raise(const ParseInput & in, States &&... st)
{
if (in.empty()) {
std::string expected;
if constexpr (constexpr auto msg = error_message<Rule>)
expected = fmt(", %s", msg);
throw p::parse_error("unexpected end of file" + expected, in);
}
p::must_if<SyntaxErrors>::control<Rule>::raise(in, st...);
}
};
struct ExprState : grammar::operator_semantics<ExprState, PosIdx, AttrPath, std::unique_ptr<Expr>>
{
template<typename Op, typename... Args>
std::unique_ptr<Expr> applyUnary(Args &&... args) {
return std::make_unique<Op>(popExpr(), std::forward<Args>(args)...);
}
template<typename Op>
std::unique_ptr<Expr> applyBinary(PosIdx pos) {
auto right = popExpr(), left = popExpr();
return std::make_unique<Op>(pos, std::move(left), std::move(right));
}
std::unique_ptr<Expr> call(PosIdx pos, Symbol fn, bool flip = false)
{
std::vector<std::unique_ptr<Expr>> args(2);
args[flip ? 0 : 1] = popExpr();
args[flip ? 1 : 0] = popExpr();
return std::make_unique<ExprCall>(pos, std::make_unique<ExprVar>(fn), std::move(args));
}
std::unique_ptr<Expr> order(PosIdx pos, bool less, State & state)
{
return call(pos, state.s.lessThan, !less);
}
std::unique_ptr<Expr> concatStrings(PosIdx pos)
{
std::vector<std::pair<PosIdx, std::unique_ptr<Expr>>> args(2);
args[1].second = popExpr();
args[0].second = popExpr();
return std::make_unique<ExprConcatStrings>(pos, false, std::move(args));
}
std::unique_ptr<Expr> negate(PosIdx pos, State & state)
{
std::vector<std::unique_ptr<Expr>> args(2);
args[0] = std::make_unique<ExprInt>(0);
args[1] = popExpr();
return std::make_unique<ExprCall>(pos, std::make_unique<ExprVar>(state.s.sub), std::move(args));
}
std::unique_ptr<Expr> applyOp(PosIdx pos, auto & op, State & state) {
using Op = grammar::op;
auto not_ = [] (auto e) {
return std::make_unique<ExprOpNot>(std::move(e));
};
return (overloaded {
[&] (Op::implies) { return applyBinary<ExprOpImpl>(pos); },
[&] (Op::or_) { return applyBinary<ExprOpOr>(pos); },
[&] (Op::and_) { return applyBinary<ExprOpAnd>(pos); },
[&] (Op::equals) { return applyBinary<ExprOpEq>(pos); },
[&] (Op::not_equals) { return applyBinary<ExprOpNEq>(pos); },
[&] (Op::less) { return order(pos, true, state); },
[&] (Op::greater_eq) { return not_(order(pos, true, state)); },
[&] (Op::greater) { return order(pos, false, state); },
[&] (Op::less_eq) { return not_(order(pos, false, state)); },
[&] (Op::update) { return applyBinary<ExprOpUpdate>(pos); },
[&] (Op::not_) { return applyUnary<ExprOpNot>(); },
[&] (Op::plus) { return concatStrings(pos); },
[&] (Op::minus) { return call(pos, state.s.sub); },
[&] (Op::mul) { return call(pos, state.s.mul); },
[&] (Op::div) { return call(pos, state.s.div); },
[&] (Op::concat) { return applyBinary<ExprOpConcatLists>(pos); },
[&] (has_attr & a) { return applyUnary<ExprOpHasAttr>(std::move(a.path)); },
[&] (Op::unary_minus) { return negate(pos, state); },
})(op);
}
// always_inline is needed, otherwise pushOp slows down considerably
[[noreturn, gnu::always_inline]]
static void badOperator(PosIdx pos, State & state)
{
throw ParseError({
.msg = HintFmt("syntax error, unexpected operator"),
.pos = state.positions[pos]
});
}
template<typename Expr, typename... Args>
Expr & pushExpr(Args && ... args)
{
auto p = std::make_unique<Expr>(std::forward<Args>(args)...);
auto & result = *p;
exprs.emplace_back(std::move(p));
return result;
}
};
struct SubexprState {
private:
ExprState * up;
public:
explicit SubexprState(ExprState & up, auto &...) : up(&up) {}
operator ExprState &() { return *up; }
ExprState * operator->() { return up; }
};
template<typename Rule>
struct BuildAST : grammar::nothing<Rule> {};
struct LambdaState : SubexprState {
using SubexprState::SubexprState;
Symbol arg;
std::unique_ptr<Formals> formals;
};
struct FormalsState : SubexprState {
using SubexprState::SubexprState;
Formals formals{};
Formal formal{};
};
template<> struct BuildAST<grammar::formal::name> {
static void apply(const auto & in, FormalsState & s, State & ps) {
s.formal = {
.pos = ps.at(in),
.name = ps.symbols.create(in.string_view()),
};
}
};
template<> struct BuildAST<grammar::formal> {
static void apply0(FormalsState & s, State &) {
s.formals.formals.emplace_back(std::move(s.formal));
}
};
template<> struct BuildAST<grammar::formal::default_value> {
static void apply0(FormalsState & s, State & ps) {
s.formal.def = s->popExpr();
}
};
template<> struct BuildAST<grammar::formals::ellipsis> {
static void apply0(FormalsState & s, State &) {
s.formals.ellipsis = true;
}
};
template<> struct BuildAST<grammar::formals> : change_head<FormalsState> {
static void success0(FormalsState & f, LambdaState & s, State &) {
s.formals = std::make_unique<Formals>(std::move(f.formals));
}
};
struct AttrState : SubexprState {
using SubexprState::SubexprState;
std::vector<AttrName> attrs;
void pushAttr(auto && attr, PosIdx) { attrs.emplace_back(std::move(attr)); }
};
template<> struct BuildAST<grammar::attr::simple> {
static void apply(const auto & in, auto & s, State & ps) {
s.pushAttr(ps.symbols.create(in.string_view()), ps.at(in));
}
};
template<> struct BuildAST<grammar::attr::string> {
static void apply(const auto & in, auto & s, State & ps) {
auto e = s->popExpr();
if (auto str = dynamic_cast<ExprString *>(e.get()))
s.pushAttr(ps.symbols.create(str->s), ps.at(in));
else
s.pushAttr(std::move(e), ps.at(in));
}
};
template<> struct BuildAST<grammar::attr::expr> : BuildAST<grammar::attr::string> {};
struct BindingsState : SubexprState {
using SubexprState::SubexprState;
ExprAttrs attrs;
AttrPath path;
std::unique_ptr<Expr> value;
};
struct InheritState : SubexprState {
using SubexprState::SubexprState;
std::vector<std::pair<AttrName, PosIdx>> attrs;
std::unique_ptr<Expr> from;
PosIdx fromPos;
void pushAttr(auto && attr, PosIdx pos) { attrs.emplace_back(std::move(attr), pos); }
};
template<> struct BuildAST<grammar::inherit::from> {
static void apply(const auto & in, InheritState & s, State & ps) {
s.from = s->popExpr();
s.fromPos = ps.at(in);
}
};
template<> struct BuildAST<grammar::inherit> : change_head<InheritState> {
static void success0(InheritState & s, BindingsState & b, State & ps) {
auto & attrs = b.attrs.attrs;
// TODO this should not reuse generic attrpath rules.
for (auto & [i, iPos] : s.attrs) {
if (i.symbol)
continue;
if (auto str = dynamic_cast<ExprString *>(i.expr.get()))
i = AttrName(ps.symbols.create(str->s));
else {
throw ParseError({
.msg = HintFmt("dynamic attributes not allowed in inherit"),
.pos = ps.positions[iPos]
});
}
}
if (auto fromE = std::move(s.from)) {
if (!b.attrs.inheritFromExprs)
b.attrs.inheritFromExprs = std::make_unique<std::vector<std::unique_ptr<Expr>>>();
b.attrs.inheritFromExprs->push_back(std::move(fromE));
for (auto & [i, iPos] : s.attrs) {
if (attrs.find(i.symbol) != attrs.end())
ps.dupAttr(i.symbol, iPos, attrs[i.symbol].pos);
auto from = std::make_unique<ExprInheritFrom>(s.fromPos, b.attrs.inheritFromExprs->size() - 1);
attrs.emplace(
i.symbol,
ExprAttrs::AttrDef(
std::make_unique<ExprSelect>(iPos, std::move(from), i.symbol),
iPos,
ExprAttrs::AttrDef::Kind::InheritedFrom));
}
} else {
for (auto & [i, iPos] : s.attrs) {
if (attrs.find(i.symbol) != attrs.end())
ps.dupAttr(i.symbol, iPos, attrs[i.symbol].pos);
attrs.emplace(
i.symbol,
ExprAttrs::AttrDef(
std::make_unique<ExprVar>(iPos, i.symbol),
iPos,
ExprAttrs::AttrDef::Kind::Inherited));
}
}
}
};
template<> struct BuildAST<grammar::binding::path> : change_head<AttrState> {
static void success0(AttrState & a, BindingsState & s, State & ps) {
s.path = std::move(a.attrs);
}
};
template<> struct BuildAST<grammar::binding::value> {
static void apply0(BindingsState & s, State & ps) {
s.value = s->popExpr();
}
};
template<> struct BuildAST<grammar::binding> {
static void apply(const auto & in, BindingsState & s, State & ps) {
ps.addAttr(&s.attrs, std::move(s.path), std::move(s.value), ps.at(in));
}
};
template<> struct BuildAST<grammar::expr::id> {
static void apply(const auto & in, ExprState & s, State & ps) {
if (in.string_view() == "__curPos")
s.pushExpr<ExprPos>(ps.at(in));
else
s.pushExpr<ExprVar>(ps.at(in), ps.symbols.create(in.string_view()));
}
};
template<> struct BuildAST<grammar::expr::int_> {
static void apply(const auto & in, ExprState & s, State & ps) {
int64_t v;
if (std::from_chars(in.begin(), in.end(), v).ec != std::errc{}) {
throw ParseError({
.msg = HintFmt("invalid integer '%1%'", in.string_view()),
.pos = ps.positions[ps.at(in)],
});
}
s.pushExpr<ExprInt>(v);
}
};
template<> struct BuildAST<grammar::expr::float_> {
static void apply(const auto & in, ExprState & s, State & ps) {
// copy the input into a temporary string so we can call stod.
// can't use from_chars because libc++ (thus darwin) does not have it,
// and floats are not performance-sensitive anyway. if they were you'd
// be in much bigger trouble than this.
//
// we also get to do a locale-save dance because stod is locale-aware and
// something (a plugin?) may have called setlocale or uselocale.
static struct locale_hack {
locale_t posix;
locale_hack(): posix(newlocale(LC_ALL_MASK, "POSIX", 0))
{
if (posix == 0)
throw SysError("could not get POSIX locale");
}
} locale;
auto tmp = in.string();
double v = [&] {
auto oldLocale = uselocale(locale.posix);
Finally resetLocale([=] { uselocale(oldLocale); });
try {
return std::stod(tmp);
} catch (...) {
throw ParseError({
.msg = HintFmt("invalid float '%1%'", in.string_view()),
.pos = ps.positions[ps.at(in)],
});
}
}();
s.pushExpr<ExprFloat>(v);
}
};
struct StringState : SubexprState {
using SubexprState::SubexprState;
std::string currentLiteral;
PosIdx currentPos;
std::vector<std::pair<nix::PosIdx, std::unique_ptr<Expr>>> parts;
void append(PosIdx pos, std::string_view s)
{
if (currentLiteral.empty())
currentPos = pos;
currentLiteral += s;
}
// FIXME this truncates strings on NUL for compat with the old parser. ideally
// we should use the decomposition the g gives us instead of iterating over
// the entire string again.
static void unescapeStr(std::string & str)
{
char * s = str.data();
char * t = s;
char c;
while ((c = *s++)) {
if (c == '\\') {
c = *s++;
if (c == 'n') *t = '\n';
else if (c == 'r') *t = '\r';
else if (c == 't') *t = '\t';
else *t = c;
}
else if (c == '\r') {
/* Normalise CR and CR/LF into LF. */
*t = '\n';
if (*s == '\n') s++; /* cr/lf */
}
else *t = c;
t++;
}
str.resize(t - str.data());
}
void endLiteral()
{
if (!currentLiteral.empty()) {
unescapeStr(currentLiteral);
parts.emplace_back(currentPos, std::make_unique<ExprString>(std::move(currentLiteral)));
}
}
std::unique_ptr<Expr> finish()
{
if (parts.empty()) {
unescapeStr(currentLiteral);
return std::make_unique<ExprString>(std::move(currentLiteral));
} else {
endLiteral();
auto pos = parts[0].first;
return std::make_unique<ExprConcatStrings>(pos, true, std::move(parts));
}
}
};
template<typename... Content> struct BuildAST<grammar::string::literal<Content...>> {
static void apply(const auto & in, StringState & s, State & ps) {
s.append(ps.at(in), in.string_view());
}
};
template<> struct BuildAST<grammar::string::cr_lf> {
static void apply(const auto & in, StringState & s, State & ps) {
s.append(ps.at(in), in.string_view()); // FIXME compat with old parser
}
};
template<> struct BuildAST<grammar::string::interpolation> {
static void apply(const auto & in, StringState & s, State & ps) {
s.endLiteral();
s.parts.emplace_back(ps.at(in), s->popExpr());
}
};
template<> struct BuildAST<grammar::string::escape> {
static void apply(const auto & in, StringState & s, State & ps) {
s.append(ps.at(in), "\\"); // FIXME compat with old parser
s.append(ps.at(in), in.string_view());
}
};
template<> struct BuildAST<grammar::string> : change_head<StringState> {
static void success0(StringState & s, ExprState & e, State &) {
e.exprs.push_back(s.finish());
}
};
struct IndStringState : SubexprState {
using SubexprState::SubexprState;
std::vector<std::pair<PosIdx, std::variant<std::unique_ptr<Expr>, StringToken>>> parts;
};
template<bool Indented, typename... Content>
struct BuildAST<grammar::ind_string::literal<Indented, Content...>> {
static void apply(const auto & in, IndStringState & s, State & ps) {
s.parts.emplace_back(ps.at(in), StringToken{in.string_view(), Indented});
}
};
template<> struct BuildAST<grammar::ind_string::interpolation> {
static void apply(const auto & in, IndStringState & s, State & ps) {
s.parts.emplace_back(ps.at(in), s->popExpr());
}
};
template<> struct BuildAST<grammar::ind_string::escape> {
static void apply(const auto & in, IndStringState & s, State & ps) {
switch (*in.begin()) {
case 'n': s.parts.emplace_back(ps.at(in), StringToken{"\n"}); break;
case 'r': s.parts.emplace_back(ps.at(in), StringToken{"\r"}); break;
case 't': s.parts.emplace_back(ps.at(in), StringToken{"\t"}); break;
default: s.parts.emplace_back(ps.at(in), StringToken{in.string_view()}); break;
}
}
};
template<> struct BuildAST<grammar::ind_string> : change_head<IndStringState> {
static void success(const auto & in, IndStringState & s, ExprState & e, State & ps) {
e.exprs.emplace_back(ps.stripIndentation(ps.at(in), std::move(s.parts)));
}
};
template<typename... Content> struct BuildAST<grammar::path::literal<Content...>> {
static void apply(const auto & in, StringState & s, State & ps) {
s.append(ps.at(in), in.string_view());
s.endLiteral();
}
};
template<> struct BuildAST<grammar::path::interpolation> : BuildAST<grammar::string::interpolation> {};
template<> struct BuildAST<grammar::path::anchor> {
static void apply(const auto & in, StringState & s, State & ps) {
Path path(absPath(in.string(), ps.basePath.path.abs()));
/* add back in the trailing '/' to the first segment */
if (in.string_view().ends_with('/') && in.size() > 1)
path += "/";
s.parts.emplace_back(ps.at(in), new ExprPath(std::move(path)));
}
};
template<> struct BuildAST<grammar::path::home_anchor> {
static void apply(const auto & in, StringState & s, State & ps) {
if (evalSettings.pureEval)
throw Error("the path '%s' can not be resolved in pure mode", in.string_view());
Path path(getHome() + in.string_view().substr(1));
s.parts.emplace_back(ps.at(in), new ExprPath(std::move(path)));
}
};
template<> struct BuildAST<grammar::path::searched_path> {
static void apply(const auto & in, StringState & s, State & ps) {
std::vector<std::unique_ptr<Expr>> args{2};
args[0] = std::make_unique<ExprVar>(ps.s.nixPath);
args[1] = std::make_unique<ExprString>(in.string());
s.parts.emplace_back(
ps.at(in),
std::make_unique<ExprCall>(
ps.at(in),
std::make_unique<ExprVar>(ps.s.findFile),
std::move(args)));
}
};
template<> struct BuildAST<grammar::path> : change_head<StringState> {
template<typename E>
static void check_slash(PosIdx end, StringState & s, State & ps) {
auto e = dynamic_cast<E *>(s.parts.back().second.get());
if (!e || !e->s.ends_with('/'))
return;
if (s.parts.size() > 1 || e->s != "/")
throw ParseError({
.msg = HintFmt("path has a trailing slash"),
.pos = ps.positions[end],
});
}
static void success(const auto & in, StringState & s, ExprState & e, State & ps) {
s.endLiteral();
check_slash<ExprPath>(ps.atEnd(in), s, ps);
check_slash<ExprString>(ps.atEnd(in), s, ps);
if (s.parts.size() == 1) {
e.exprs.emplace_back(std::move(s.parts.back().second));
} else {
e.pushExpr<ExprConcatStrings>(ps.at(in), false, std::move(s.parts));
}
}
};
// strings and paths sare handled fully by the grammar-level rule for now
template<> struct BuildAST<grammar::expr::string> : p::maybe_nothing {};
template<> struct BuildAST<grammar::expr::ind_string> : p::maybe_nothing {};
template<> struct BuildAST<grammar::expr::path> : p::maybe_nothing {};
template<> struct BuildAST<grammar::expr::uri> {
static void apply(const auto & in, ExprState & s, State & ps) {
static bool noURLLiterals = experimentalFeatureSettings.isEnabled(Xp::NoUrlLiterals);
if (noURLLiterals)
throw ParseError({
.msg = HintFmt("URL literals are disabled"),
.pos = ps.positions[ps.at(in)]
});
s.pushExpr<ExprString>(in.string());
}
};
template<> struct BuildAST<grammar::expr::ancient_let> : change_head<BindingsState> {
static void success(const auto & in, BindingsState & b, ExprState & s, State & ps) {
b.attrs.pos = ps.at(in);
b.attrs.recursive = true;
s.pushExpr<ExprSelect>(b.attrs.pos, std::make_unique<ExprAttrs>(std::move(b.attrs)), ps.s.body);
}
};
template<> struct BuildAST<grammar::expr::rec_set> : change_head<BindingsState> {
static void success(const auto & in, BindingsState & b, ExprState & s, State & ps) {
b.attrs.pos = ps.at(in);
b.attrs.recursive = true;
s.pushExpr<ExprAttrs>(std::move(b.attrs));
}
};
template<> struct BuildAST<grammar::expr::set> : change_head<BindingsState> {
static void success(const auto & in, BindingsState & b, ExprState & s, State & ps) {
b.attrs.pos = ps.at(in);
s.pushExpr<ExprAttrs>(std::move(b.attrs));
}
};
using ListState = std::vector<std::unique_ptr<Expr>>;
template<> struct BuildAST<grammar::expr::list> : change_head<ListState> {
static void success0(ListState & ls, ExprState & s, State &) {
auto e = std::make_unique<ExprList>();
e->elems = std::move(ls);
s.exprs.push_back(std::move(e));
}
};
template<> struct BuildAST<grammar::expr::list::entry> : change_head<ExprState> {
static void success0(ExprState & e, ListState & s, State & ps) {
s.emplace_back(e.finish(ps));
}
};
struct SelectState : SubexprState {
using SubexprState::SubexprState;
PosIdx pos;
ExprSelect * e = nullptr;
};
template<> struct BuildAST<grammar::expr::select::head> {
static void apply(const auto & in, SelectState & s, State & ps) {
s.pos = ps.at(in);
}
};
template<> struct BuildAST<grammar::expr::select::attr> : change_head<AttrState> {
static void success0(AttrState & a, SelectState & s, State &) {
s.e = &s->pushExpr<ExprSelect>(s.pos, s->popExpr(), std::move(a.attrs), nullptr);
}
};
template<> struct BuildAST<grammar::expr::select::attr_or> {
static void apply0(SelectState & s, State &) {
s.e->def = s->popExpr();
}
};
template<> struct BuildAST<grammar::expr::select::as_app_or> {
static void apply(const auto & in, SelectState & s, State & ps) {
std::vector<std::unique_ptr<Expr>> args(1);
args[0] = std::make_unique<ExprVar>(ps.at(in), ps.s.or_);
s->pushExpr<ExprCall>(s.pos, s->popExpr(), std::move(args));
}
};
template<> struct BuildAST<grammar::expr::select> : change_head<SelectState> {
static void success0(const auto &...) {}
};
struct AppState : SubexprState {
using SubexprState::SubexprState;
PosIdx pos;
ExprCall * e = nullptr;
};
template<> struct BuildAST<grammar::expr::app::select_or_fn> {
static void apply(const auto & in, AppState & s, State & ps) {
s.pos = ps.at(in);
}
};
template<> struct BuildAST<grammar::expr::app::first_arg> {
static void apply(auto & in, AppState & s, State & ps) {
auto arg = s->popExpr(), fn = s->popExpr();
if ((s.e = dynamic_cast<ExprCall *>(fn.get()))) {
// TODO remove.
// AST compat with old parser, semantics are the same.
// this can happen on occasions such as `<p> <p>` or `a or b or`,
// neither of which are super worth optimizing.
s.e->args.push_back(std::move(arg));
s->exprs.emplace_back(std::move(fn));
} else {
std::vector<std::unique_ptr<Expr>> args{1};
args[0] = std::move(arg);
s.e = &s->pushExpr<ExprCall>(s.pos, std::move(fn), std::move(args));
}
}
};
template<> struct BuildAST<grammar::expr::app::another_arg> {
static void apply0(AppState & s, State & ps) {
s.e->args.push_back(s->popExpr());
}
};
template<> struct BuildAST<grammar::expr::app> : change_head<AppState> {
static void success0(const auto &...) {}
};
template<typename Op> struct BuildAST<grammar::expr::operator_<Op>> {
static void apply(const auto & in, ExprState & s, State & ps) {
s.pushOp(ps.at(in), Op{}, ps);
}
};
template<> struct BuildAST<grammar::expr::operator_<grammar::op::has_attr>> : change_head<AttrState> {
static void success(const auto & in, AttrState & a, ExprState & s, State & ps) {
s.pushOp(ps.at(in), ExprState::has_attr{{}, std::move(a.attrs)}, ps);
}
};
template<> struct BuildAST<grammar::expr::lambda::arg> {
static void apply(const auto & in, LambdaState & s, State & ps) {
s.arg = ps.symbols.create(in.string_view());
}
};
template<> struct BuildAST<grammar::expr::lambda> : change_head<LambdaState> {
static void success(const auto & in, LambdaState & l, ExprState & s, State & ps) {
if (l.formals)
l.formals = ps.validateFormals(std::move(l.formals), ps.at(in), l.arg);
s.pushExpr<ExprLambda>(ps.at(in), l.arg, std::move(l.formals), l->popExpr());
}
};
template<> struct BuildAST<grammar::expr::assert_> {
static void apply(const auto & in, ExprState & s, State & ps) {
auto body = s.popExpr(), cond = s.popExpr();
s.pushExpr<ExprAssert>(ps.at(in), std::move(cond), std::move(body));
}
};
template<> struct BuildAST<grammar::expr::with> {
static void apply(const auto & in, ExprState & s, State & ps) {
auto body = s.popExpr(), scope = s.popExpr();
s.pushExpr<ExprWith>(ps.at(in), std::move(scope), std::move(body));
}
};
template<> struct BuildAST<grammar::expr::let> : change_head<BindingsState> {
static void success(const auto & in, BindingsState & b, ExprState & s, State & ps) {
if (!b.attrs.dynamicAttrs.empty())
throw ParseError({
.msg = HintFmt("dynamic attributes not allowed in let"),
.pos = ps.positions[ps.at(in)]
});
s.pushExpr<ExprLet>(std::make_unique<ExprAttrs>(std::move(b.attrs)), b->popExpr());
}
};
template<> struct BuildAST<grammar::expr::if_> {
static void apply(const auto & in, ExprState & s, State & ps) {
auto else_ = s.popExpr(), then = s.popExpr(), cond = s.popExpr();
s.pushExpr<ExprIf>(ps.at(in), std::move(cond), std::move(then), std::move(else_));
}
};
template<> struct BuildAST<grammar::expr> : change_head<ExprState> {
static void success0(ExprState & inner, ExprState & outer, State & ps) {
outer.exprs.push_back(inner.finish(ps));
}
};
}
}
namespace nix {
Expr * EvalState::parse(
char * text,
size_t length,
Pos::Origin origin,
const SourcePath & basePath,
std::shared_ptr<StaticEnv> & staticEnv)
{
parser::State s = {
symbols,
positions,
basePath,
positions.addOrigin(origin, length),
exprSymbols,
};
parser::ExprState x;
assert(length >= 2);
assert(text[length - 1] == 0);
assert(text[length - 2] == 0);
length -= 2;
p::string_input<p::tracking_mode::lazy> inp{std::string_view{text, length}, "input"};
try {
p::parse<parser::grammar::root, parser::BuildAST, parser::Control>(inp, x, s);
} catch (p::parse_error & e) {
auto pos = e.positions().back();
throw ParseError({
.msg = HintFmt("syntax error, %s", e.message()),
.pos = positions[s.positions.add(s.origin, pos.byte)]
});
}
auto result = x.finish(s);
result->bindVars(*this, staticEnv);
return result.release();
}
}

View file

@ -3,77 +3,61 @@
#include "eval.hh"
namespace nix {
namespace nix::parser {
/**
* @note Storing a C-style `char *` and `size_t` allows us to avoid
* having to define the special members that using string_view here
* would implicitly delete.
*/
struct StringToken
{
const char * p;
size_t l;
std::string_view s;
bool hasIndentation;
operator std::string_view() const { return {p, l}; }
operator std::string_view() const { return s; }
};
struct ParserLocation
{
int first_line, first_column;
int last_line, last_column;
// backup to recover from yyless(0)
int stashed_first_column, stashed_last_column;
void stash() {
stashed_first_column = first_column;
stashed_last_column = last_column;
}
void unstash() {
first_column = stashed_first_column;
last_column = stashed_last_column;
}
};
struct ParserState
struct State
{
SymbolTable & symbols;
PosTable & positions;
Expr * result;
SourcePath basePath;
PosTable::Origin origin;
const Expr::AstSymbols & s;
std::unique_ptr<Error> error;
[[nodiscard]] ParseError dupAttr(const AttrPath & attrPath, const PosIdx pos, const PosIdx prevPos);
[[nodiscard]] ParseError dupAttr(Symbol attr, const PosIdx pos, const PosIdx prevPos);
[[nodiscard]] std::optional<ParseError> addAttr(ExprAttrs * attrs, AttrPath && attrPath, std::unique_ptr<Expr> e, const PosIdx pos);
[[nodiscard]] std::optional<ParseError> validateFormals(Formals * formals, PosIdx pos = noPos, Symbol arg = {});
void dupAttr(const AttrPath & attrPath, const PosIdx pos, const PosIdx prevPos);
void dupAttr(Symbol attr, const PosIdx pos, const PosIdx prevPos);
void addAttr(ExprAttrs * attrs, AttrPath && attrPath, std::unique_ptr<Expr> e, const PosIdx pos);
std::unique_ptr<Formals> validateFormals(std::unique_ptr<Formals> formals, PosIdx pos = noPos, Symbol arg = {});
std::unique_ptr<Expr> stripIndentation(const PosIdx pos,
std::vector<std::pair<PosIdx, std::variant<std::unique_ptr<Expr>, StringToken>>> && es);
PosIdx at(const ParserLocation & loc);
// lazy positioning means we don't get byte offsets directly, in.position() would work
// but also requires line and column (which is expensive)
PosIdx at(const auto & in)
{
return positions.add(origin, in.begin() - in.input().begin());
}
PosIdx atEnd(const auto & in)
{
return positions.add(origin, in.end() - in.input().begin());
}
};
inline ParseError ParserState::dupAttr(const AttrPath & attrPath, const PosIdx pos, const PosIdx prevPos)
inline void State::dupAttr(const AttrPath & attrPath, const PosIdx pos, const PosIdx prevPos)
{
return ParseError({
throw ParseError({
.msg = HintFmt("attribute '%1%' already defined at %2%",
showAttrPath(symbols, attrPath), positions[prevPos]),
.pos = positions[pos]
});
}
inline ParseError ParserState::dupAttr(Symbol attr, const PosIdx pos, const PosIdx prevPos)
inline void State::dupAttr(Symbol attr, const PosIdx pos, const PosIdx prevPos)
{
return ParseError({
throw ParseError({
.msg = HintFmt("attribute '%1%' already defined at %2%", symbols[attr], positions[prevPos]),
.pos = positions[pos]
});
}
inline std::optional<ParseError> ParserState::addAttr(ExprAttrs * attrs, AttrPath && attrPath, std::unique_ptr<Expr> e, const PosIdx pos)
inline void State::addAttr(ExprAttrs * attrs, AttrPath && attrPath, std::unique_ptr<Expr> e, const PosIdx pos)
{
AttrPath::iterator i;
// All attrpaths have at least one attr
@ -86,10 +70,10 @@ inline std::optional<ParseError> ParserState::addAttr(ExprAttrs * attrs, AttrPat
if (j != attrs->attrs.end()) {
if (j->second.kind != ExprAttrs::AttrDef::Kind::Inherited) {
ExprAttrs * attrs2 = dynamic_cast<ExprAttrs *>(j->second.e.get());
if (!attrs2) return dupAttr(attrPath, pos, j->second.pos);
if (!attrs2) dupAttr(attrPath, pos, j->second.pos);
attrs = attrs2;
} else
return dupAttr(attrPath, pos, j->second.pos);
dupAttr(attrPath, pos, j->second.pos);
} else {
auto next = attrs->attrs.emplace(std::piecewise_construct,
std::tuple(i->symbol),
@ -130,7 +114,7 @@ inline std::optional<ParseError> ParserState::addAttr(ExprAttrs * attrs, AttrPat
if (ae->inheritFromExprs)
std::ranges::move(*ae->inheritFromExprs, std::back_inserter(*jAttrs->inheritFromExprs));
} else {
return dupAttr(attrPath, pos, j->second.pos);
dupAttr(attrPath, pos, j->second.pos);
}
} else {
// This attr path is not defined. Let's create it.
@ -142,11 +126,9 @@ inline std::optional<ParseError> ParserState::addAttr(ExprAttrs * attrs, AttrPat
} else {
attrs->dynamicAttrs.emplace_back(std::move(i->expr), std::move(e), pos);
}
return {};
}
inline std::optional<ParseError> ParserState::validateFormals(Formals * formals, PosIdx pos, Symbol arg)
inline std::unique_ptr<Formals> State::validateFormals(std::unique_ptr<Formals> formals, PosIdx pos, Symbol arg)
{
std::sort(formals->formals.begin(), formals->formals.end(),
[] (const auto & a, const auto & b) {
@ -161,21 +143,21 @@ inline std::optional<ParseError> ParserState::validateFormals(Formals * formals,
duplicate = std::min(thisDup, duplicate.value_or(thisDup));
}
if (duplicate)
return ParseError({
throw ParseError({
.msg = HintFmt("duplicate formal function argument '%1%'", symbols[duplicate->first]),
.pos = positions[duplicate->second]
});
if (arg && formals->has(arg))
return ParseError({
throw ParseError({
.msg = HintFmt("duplicate formal function argument '%1%'", symbols[arg]),
.pos = positions[pos]
});
return {};
return formals;
}
inline std::unique_ptr<Expr> ParserState::stripIndentation(const PosIdx pos,
inline std::unique_ptr<Expr> State::stripIndentation(const PosIdx pos,
std::vector<std::pair<PosIdx, std::variant<std::unique_ptr<Expr>, StringToken>>> && es)
{
if (es.empty()) return std::make_unique<ExprString>("");
@ -196,11 +178,11 @@ inline std::unique_ptr<Expr> ParserState::stripIndentation(const PosIdx pos,
}
continue;
}
for (size_t j = 0; j < str->l; ++j) {
for (size_t j = 0; j < str->s.size(); ++j) {
if (atStartOfLine) {
if (str->p[j] == ' ')
if (str->s[j] == ' ')
curIndent++;
else if (str->p[j] == '\n') {
else if (str->s[j] == '\n') {
/* Empty line, doesn't influence minimum
indentation. */
curIndent = 0;
@ -208,7 +190,7 @@ inline std::unique_ptr<Expr> ParserState::stripIndentation(const PosIdx pos,
atStartOfLine = false;
if (curIndent < minIndent) minIndent = curIndent;
}
} else if (str->p[j] == '\n') {
} else if (str->s[j] == '\n') {
atStartOfLine = true;
curIndent = 0;
}
@ -228,23 +210,23 @@ inline std::unique_ptr<Expr> ParserState::stripIndentation(const PosIdx pos,
};
const auto trimString = [&] (const StringToken & t) {
std::string s2;
for (size_t j = 0; j < t.l; ++j) {
for (size_t j = 0; j < t.s.size(); ++j) {
if (atStartOfLine) {
if (t.p[j] == ' ') {
if (t.s[j] == ' ') {
if (curDropped++ >= minIndent)
s2 += t.p[j];
s2 += t.s[j];
}
else if (t.p[j] == '\n') {
else if (t.s[j] == '\n') {
curDropped = 0;
s2 += t.p[j];
s2 += t.s[j];
} else {
atStartOfLine = false;
curDropped = 0;
s2 += t.p[j];
s2 += t.s[j];
}
} else {
s2 += t.p[j];
if (t.p[j] == '\n') atStartOfLine = true;
s2 += t.s[j];
if (t.s[j] == '\n') atStartOfLine = true;
}
}
@ -269,9 +251,4 @@ inline std::unique_ptr<Expr> ParserState::stripIndentation(const PosIdx pos,
return std::make_unique<ExprConcatStrings>(pos, true, std::move(es2));
}
inline PosIdx ParserState::at(const ParserLocation & loc)
{
return positions.add(origin, loc.first_column);
}
}

View file

@ -1,4 +1,4 @@
error: syntax error, unexpected end of file
error: syntax error, unexpected end of file, expecting expression
at «stdin»:3:1:
2| # no content
3|

View file

@ -1,4 +1,4 @@
error: syntax error, unexpected ':', expecting '}'
error: syntax error, expecting '}'
at «stdin»:3:13:
2|
3| f = {x, y : ["baz" "bar" z "bat"]}: x + y;

View file

@ -1,4 +1,4 @@
error: syntax error, unexpected invalid token, expecting end of file
error: syntax error, expecting end of file
at «stdin»:1:5:
1| 123 é 4
| ^