https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108632
Bug ID: 108632
Summary: [13 Regression] libstdc++ std/time/hh_mm_ss/1.cc on
"packed" platforms
Product: gcc
Version: 13.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: testsuite
Assignee: unassigned at gcc dot gnu.org
Reporter: hp at gcc dot gnu.org
Target Milestone: ---
With a commit in the series r13-4522-gba1536dac780f3..r13-4532-gda7fb32d403759,
the test std/time/hh_mm_ss/1.cc started to fail for cris-elf, where the ABI
mandates structure layout that virtually corresponds to as if declared with the
attribute packed on other platforms.
In libstdc++.log:
/x/gcc/libstdc++-v3/testsuite/std/time/hh_mm_ss/1.cc: In function 'constexpr
void size()':
/x/gcc/libstdc++-v3/testsuite/std/time/hh_mm_ss/1.cc:107: error: static
assertion failed
/x/gcc/libstdc++-v3/testsuite/std/time/hh_mm_ss/1.cc:107: note: the comparison
reduces to '(12 == 11)'
Undoubtedly the cause is commit r13-4526-g5329e1a8e1480d, which included:
106: + struct S0 { long long h; char m; char s; bool neg; };
107: + static_assert(sizeof(hh_mm_ss<seconds>) == sizeof(S0));
108: + struct S1 { long long h; char m; char s; bool neg; char ss; };
109: + static_assert(sizeof(hh_mm_ss<duration<int, std::centi>>) ==
sizeof(S1));
The test appears to assert that the footprint of hh_mm_ss<seconds> is minimal,
but for that, S0 misses a trailing empty member for hh_mm_ss<seconds> , which
for most targets is hidden in padding, but which require at least one byte of
storage and thus for such targets you'll have sizeof(hh_mm_ss<seconds>) 8+1+1+1
+1 = 12.
