[PATCH] D46439: Fix incorrect packed aligned structure layout
Momchil Velikov via Phabricator via cfe-commits
cfe-commits at lists.llvm.org
Fri May 4 09:45:47 PDT 2018
chill created this revision.
chill added reviewers: rsmith, aaron.ballman.
Executing the following program
#include <cassert>
#include <cstddef>
struct S {
char x;
int y;
} __attribute__((packed, aligned(8)));
struct alignas(8) T {
char x;
int y;
} __attribute__((packed));
int main() {
assert(offsetof(S, x) == 0);
assert(offsetof(S, y) == 1);
assert(offsetof(T, x) == 0);
assert(offsetof(T, y) == 1);
}
fails with assertion
a.out: a.cc:19: int main(): Assertion `offsetof(T, y) == 1' failed.
The layout if `T` is incorrect, because it's computed and effectively cached when checking for `alignas` under-aligning the structure, however, this happens before `__attribute__((packed))` is processed.
This patch moves the processing of attributes before the `alignas` under-alignment check.
Repository:
rC Clang
https://reviews.llvm.org/D46439
Files:
lib/Sema/SemaDecl.cpp
Index: lib/Sema/SemaDecl.cpp
===================================================================
--- lib/Sema/SemaDecl.cpp
+++ lib/Sema/SemaDecl.cpp
@@ -15573,6 +15573,10 @@
if (!Completed)
Record->completeDefinition();
+ // Handle attributes before checking for alignas underalignment.
+ if (Attr)
+ ProcessDeclAttributeList(S, Record, Attr);
+
// We may have deferred checking for a deleted destructor. Check now.
if (CXXRecordDecl *CXXRecord = dyn_cast<CXXRecordDecl>(Record)) {
auto *Dtor = CXXRecord->getDestructor();
@@ -15703,9 +15707,6 @@
CDecl->setIvarRBraceLoc(RBrac);
}
}
-
- if (Attr)
- ProcessDeclAttributeList(S, Record, Attr);
}
/// \brief Determine whether the given integral value is representable within
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D46439.145213.patch
Type: text/x-patch
Size: 787 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20180504/694f7d35/attachment.bin>
More information about the cfe-commits
mailing list