This switches `DCHECK` and `FATAL` as well. The goal is to reduce the code size impact of these assertions so that we can keep more of them enabled. Currently, the largest cost I see from `CHECK` is not the actual check or the cold code itself, but actually the failure to inline trivial functions due to the presence of the cold code. This means that our goal isn't to reduce apparent code size in the final binary but the LLVM IR cost assessed for these routines in the inliner, which closely correlates with code size but is a bit different. As discussed in #4283, experimentation shows that a single function call with a minimal number of arguments is the lowest cost model for these. This is easily achieved with a format-string API that internally uses `llvm::formatv`. This PR is essentially the `CHECK` version of #4283. However, the check macros are substantially harder to make work with both format strings and streaming because they also take a condition. Also, unexpectedly, I was very successful at devising a regular expression based automated rewrite from the streaming to the format string form with only low 10s of manual fixes. This includes compacting strings broken up across lines, etc. Given how well that went, I've prepared this PR which just directly switches to the format string API and migrate everything to use it. One nice side-effect is that the format string approach ends up greatly simplifying the implementation here as well. This is ... *shockingly* effective. Parsing speeds up by more than 3% with just this change. And checking speeds up by **8%** with this change alone: ``` BM_CompileAPIFileDenseDecls<Phase::Parse>/256 86.3µs ± 1% 82.9µs ± 1% -3.94% (p=0.000 n=17+19) BM_CompileAPIFileDenseDecls<Phase::Parse>/1024 431µs ± 1% 415µs ± 1% -3.76% (p=0.000 n=18+19) BM_CompileAPIFileDenseDecls<Phase::Parse>/4096 1.77ms ± 1% 1.71ms ± 1% -3.18% (p=0.000 n=18+19) BM_CompileAPIFileDenseDecls<Phase::Parse>/16384 7.44ms ± 1% 7.17ms ± 2% -3.56% (p=0.000 n=18+20) BM_CompileAPIFileDenseDecls<Phase::Parse>/65536 30.7ms ± 1% 29.7ms ± 1% -3.15% (p=0.000 n=18+20) BM_CompileAPIFileDenseDecls<Phase::Parse>/262144 131ms ± 1% 127ms ± 1% -2.81% (p=0.000 n=18+18) BM_CompileAPIFileDenseDecls<Phase::Check>/256 878µs ± 2% 800µs ± 1% -8.91% (p=0.000 n=19+20) BM_CompileAPIFileDenseDecls<Phase::Check>/1024 1.88ms ± 2% 1.72ms ± 1% -8.56% (p=0.000 n=19+20) BM_CompileAPIFileDenseDecls<Phase::Check>/4096 5.78ms ± 2% 5.28ms ± 1% -8.70% (p=0.000 n=20+18) BM_CompileAPIFileDenseDecls<Phase::Check>/16384 21.9ms ± 1% 20.1ms ± 1% -8.02% (p=0.000 n=18+20) BM_CompileAPIFileDenseDecls<Phase::Check>/65536 90.4ms ± 2% 83.1ms ± 1% -8.04% (p=0.000 n=19+20) BM_CompileAPIFileDenseDecls<Phase::Check>/262144 381ms ± 2% 352ms ± 1% -7.79% (p=0.000 n=19+19) ``` --------- Co-authored-by: Richard Smith <richard@metafoo.co.uk> Co-authored-by: josh11b <15258583+josh11b@users.noreply.github.com>
Explorer Syntax
The code in this directory is responsible for translating Carbon source code to
the AST defined in ast. It consists primarily of a Flex lexer
defined in lexer.lpp and a Bison grammar defined in
parser.ypp.
It is possible to define and test a new expression syntax without defining its
semantics by using the UnimplementedExpression AST node type and the same
techniques can be applied to other kinds of AST nodes as needed. See the
handling of the UNIMPL_EXAMPLE token for an example of how this is done, and
see unimplemented_example_test.cpp for an
example of how to test it.
Precedence and associativity
The Bison expression grammar uses the precedence climbing method to model precedence and associativity, suitably modified to handle Carbon's partial precedence order without grammar ambiguities.
Consider this example precedence diagram:
graph BT
%%{init: {'themeVariables': {'fontFamily': 'monospace'}}}%%
minus["minus<br>-x"]
mul>"mul<br>x * y"]
add>"add<br>x + y"]
mod["mod<br>x % y"]
eq["eq<br>x = y"]
eq --> add & mod
add --> mul
mul & mod --> minus
For each precedence level, we have up to three grammar productions:
foo_expressionrepresents an expression at that precedence level or higher, and includes as productions all of the expression kinds that are immediately higher in the precedence graph:add_expression: mul_expression | add_lhs '+' add_operand ;foo_operandrepresents an operand of afoo_expressionthat is not itself afoo_expression.eq_operand: add_expression | mod_expression ;- For left-associative operators,
foo_lhsrepresents either afoo_operandor afoo_expression.add_lhs: add_operand | add_expression ;
The above approach leads to (benign) reduce-reduce conflicts. In our example
precedence diagram, the expression -x == y has two different parses:
- eq_expression
- eq_operand
- add_expression
- mul_expression
- minus_expression
-x
- minus_expression
- mul_expression
- add_expression
==- eq_operand
- ...
y
- ...
- eq_operand
and
- eq_expression
- eq_operand
- mod_expression
- minus_expression
-x
- minus_expression
- mod_expression
==- eq_operand
- ...
y
- ...
- eq_operand
These would invoke the same parsing actions, so the states can be combined, but Bison isn't smart enough to see that.
In order to eliminate these conflicts, if there are multiple paths through the
precedence graph between a higher-precedence level foo and some lower
precedence level bar -- that is, if there's a diamond in the precedence graph
with foo at the top and bar at the bottom -- foo_expressions are excluded
from all intermediate _expression productions on the diamond between foo and
bar, and are added back in the downstream _operand productions in the
diamond instead:
minus_expression:
identifier | '-' identifier ;
// In the real grammar, trivial productions like this are inlined.
mul_operand:
minus_expression ;
mul_lhs:
mul_operand | mul_expression ;
// A minus_expression is not a mul_expression, even though it's a
// higher-precedence expression, because there are multiple paths from
// eq_expression to minus_expression, and this production is on such a path.
mul_expression:
mul_lhs '*' mul_operand
// minus_expression is listed here because it is excluded from mul_expression.
add_operand:
minus_expression | mul_expression ;
// This is notionally
// add_operand | add_expression
// but that introduces another kind of reduce-reduce conflict, because there
// would be two ways to interpret a mul_expression as an add_lhs.
add_lhs:
minus_expression | add_expression ;
// A mul_expression is an add_expression, because multiplication is
// higher-precedence, and mul is not at the top of a diamond in the precedence
// graph. minus_expression is excluded because we are within a diamond with it
// at the top.
add_expression:
mul_expression | add_lhs '+' add_operand ;
mod_operand:
minus_expression ;
mod_expression:
mod_operand '%' mod_operand ;
// We add back minus_expression here because it was excluded from add_expression
// and mod_expression.
eq_operand:
minus_expression | add_expression | mod_expression ;
// We also include minus_expression here because this is the bottom of the
// precedence diamond.
eq_expression:
minus_expression | add_expression | mod_expression | eq_operand '=' eq_operand ;