6 Commits

Author SHA1 Message Date
Aaron Ballman
2f48ca3aec Revert "Silence MSVC warnings; NFC"
This reverts commit 44c0719e77b37374c89b7fc1320664ebb404323d.

It broke several -Werror bots because of misuse of override.
2025-01-30 12:23:18 -05:00
Aaron Ballman
44c0719e77 Silence MSVC warnings; NFC
After the changes to DynamicRecursiveASTVisitor, we started getting
several warnings from MSVC like:

warning C4661: 'bool clang::DynamicRecursiveASTVisitorBase<false>::WalkUpFromNamedDecl(clang::NamedDecl *)':
no suitable definition provided for explicit template instantiation request

These changes silence the warnings by providing a definition for those
functions.
2025-01-30 12:10:48 -05:00
Sirraide
39a72be5e5
[Clang] [NFC] Introduce ConstDynamicRecursiveASTVisitor (reland) (#124821)
This relands #122991 (eeefa72).

The last attempt at landing this caused some problems; I’m not entirely
sure what happened, but it might have been due to an unnecessary use
of the `template` keyword in a few places. This removes that and attempts
to land the change again.
2025-01-29 16:14:18 +01:00
Sirraide
7873d3b50b
Revert "[Clang] [NFC] Introduce ConstDynamicRecursiveASTVisitor" (#124667)
Reverts llvm/llvm-project#122991

One of the bots is breaking; I’ll have to investigate what the issue is;
this might be because I haven’t updated the branch in a while.
2025-01-28 02:21:18 +01:00
Sirraide
eeefa72f7c
[Clang] [NFC] Introduce ConstDynamicRecursiveASTVisitor (#122991)
After some discussion around #116823, it was decided that it would be
nice to have a `const` variant of `DynamicRecursiveASTVisitor`, so this
pr does exactly that by making the main DRAV implementation a template
with a single `bool` template parameter that turns several function
parameters from a `T*` or `T&` to a `const T*` or `const T&`.

Since that made the implementation of a bunch of DRAV functions quite a
bit more verbose, I’ve moved most of them to be stamped out by a macro,
which imo makes it easier to understand what’s actually going on there.

For functions which already accepted `const` parameters in the original
RAV implementation, the parameter is `const` in both versions (e.g.
`TraverseTemplateArgument()` always takes a `const TemplateArgument&`);
conversely, parameters that are passed by value (e.g. in
`TraverseType()`, which takes a `QualType` by value) are *not* `const`
in either variant (i.e. the `QualType` argument is always just a
`QualType`, never a `const QualType`).

As a sanity check, I’ve also migrated some random visitor in the static
analyser to the `const` version (and indeed, it ends up simplifying the
code around that particular visitor actually). It would make sense to do
a pass over all visitors and change all which can be `const` use the
`const` version, but that can be done in a follow-up pr.

The [performance
impact](https://llvm-compile-time-tracker.com/compare.php?from=e3cd88a7be1dfd912bb6e7c7e888e7b442ffb5de&to=d55c5afe4a485b6d0431386e6f45cb44c1fc8883&stat=instructions:u)
of this change seems to be negligible. Clang’s binary size went up by
0.5%, but that’s expected considering that this effectively adds an
extra instantiation of `RecursiveASTVisitor`. Fortunately, this is of
course a one-time cost.
2025-01-28 02:15:02 +01:00
Sirraide
ff5551cdb0
[Clang] [NFC] Introduce DynamicRecursiveASTVisitor (#110040)
See #105195 as well as the big comment in DynamicRecursiveASTVisitor.cpp
for more context.
2024-11-05 19:23:44 +01:00