[clang-repl] Emit const variables only once #65257
Conversation
hahnjo
added
the
clang:frontend
I'd say we should not backport to 17.0.0; we need to get the final rc out the door and this isn't fixing a critical bug or regression. However, it may be reasonable to consider for 17.0.1+ |
Disable internal linkage for const variables if IncrementalExtensions are enabled. Otherwise the variables are emitted multiple times, with multiple constructions at unique memory locations, during every PTU.
hahnjo
force-pushed
the
clang-repl-const
branch
from
fcea576
to
7b52d2a
Compare
|
The original patch worked for |
|
Ping on the updated patch, which is now also passing downstream testing after I realized that we were missing a backport of commit e451d55, which explains why some constructors were wrongly ordered... |
|
ping @vgvassilev |
| // RUN: cat %s | clang-repl -Xcc -O2 | FileCheck %s | ||
|
|
||
| extern "C" int printf(const char*, ...); | ||
|
|
There was a problem hiding this comment.
Can you add a case for a file scope constant? We also should consider enabling that for C.
There was a problem hiding this comment.
const A a(1); is a file-scope constant, no? We don't need it for C because the special case in LinkageComputer::getLVForNamespaceScopeDecl only applies to C++ (my understanding is that const variables always have an identity in C).
There was a problem hiding this comment.
const A a(1); is a file-scope constant, no?
Yes, missed that.
| // RUN: cat %s | clang-repl -Xcc -O2 | FileCheck %s | ||
|
|
||
| extern "C" int printf(const char*, ...); | ||
|
|
There was a problem hiding this comment.
const A a(1); is a file-scope constant, no?
Yes, missed that.
|
Note that this currently doesn't seem to work on Windows: #68092 |
Disable internal linkage for const variables if IncrementalExtensions are enabled. Otherwise the variables are emitted multiple times, with multiple constructions at unique memory locations, during every PTU.