hokein added a comment.
> Would it be possible to compile some big file in LLVM (probably doesn't
> matter much which, Sema something?) and observe if there's a significant
> change in overall ASTContext size?
~3% saving (measuring the `ASTContext::.getASTAllocatedMemory`)
| | Before | After |
| SemaDecl.cpp | 255.5 MB | 247.5MB |
| SemaExpr.cpp | 293.5 MB | 283.5MB |
================
Comment at: clang/include/clang/AST/TemplateBase.h:43
+// the dependency.
+template <> struct PointerLikeTypeTraits<clang::Expr *> {
+ static inline void *getAsVoidPointer(clang::Expr *P) { return P; }
----------------
sammccall wrote:
> At first glance this is unsafe: you could have two different definitions of
> the same specialization in different files.
>
> In fact it's OK, the default one can now never be instantiated, because
> Expr.h includes this file and so anyone that can see the definition can see
> this specialization.
>
> But this is subtle/fragile: at least it needs to be spelled out explicitly in
> the comment.
> I'd also suggest adding a static assert below the definition of Expr that it
> is compatible with this specialization (because it is sufficiently aligned).
>
> (I can't think of a better alternative - use of PointerUnion is a win, partly
> *because* it validates the alignment)
yes, exactly.
do you have a better idea on how the static_assert should look like? The way I
can think of is to add a new flag in this specialization, and use
`static_assert(PointerLikeTypeTraits<clang::Expr *>::flag, "...")`.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D87080/new/
https://reviews.llvm.org/D87080
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
