This issue is closely related to #61144, but seems to have appeared earlier. Consider the following test case:
static int dummy = 0;
extern int foo __attribute__((__weak__, __alias__("dummy")));
typedef char check[2*!__builtin_constant_p(dummy)-1];
typedef char check[2*!__builtin_constant_p(foo)-1];
While the static dummy could be considered constant if its address did not leak outside the translation unit, its address DOES leak via the weak alias foo. As long as there is no strong definition for foo in another translation unit, dummy may be externally modified via the alias foo. As for foo, it is external and thus may never be considered constant. Thus, __builtin_constant_p(dummy) and __builtin_constant_p(foo) should both yield zero and the above translation unit should compile successfully.
The results observed vary highly with gcc version. A number of versions were observed giving __builtin_constant_p(dummy)==1, including 4.8.1, but __builtin_constant_p(foo) was more rare, showing up only in 4.9.0 and some really old versions like 4.4.7 (tested via gcc.godbolt.org since I don't have that many gcc versions lying around to test locally).
I should amend this report with further info: __builtin_constant_p is giving different results for dummy and foo depending on whether it's used at file scope or in a function. I've also observed some other really weird behavior from it that I need to summarize and report.
I think this has been fixed with last alias reorg, but double-checking and having a testcase would be nice.
Date: Sun Sep 11 12:15:02 2016
New Revision: 240082
* compile/pr61159.c: New testcase
The new testcase FAILs on Solaris/x86 with /bin/as:
FAIL: gcc.c-torture/compile/pr61159.c -O0 (test for excess errors)
"/var/tmp//ccRtFPva.s", line 5 : Cannot set a weak symbol to a common symbol
Can the bug be marked as resolved?
> --- Comment #6 from Martin Liška <marxin at gcc dot gnu.org> ---
> Can the bug be marked as resolved?
No, the Solaris/x86 problem persists.