This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: darwin8/-m64 make check failure
On 11 Jun 2008, at 09:28, Paolo Carlini wrote:
... by the way, if the issue affects only darwin8 and cannot be
reproduced on darwin9, I have vague recollections that can well be
due to a bug of the darwin8 libc. In that case we should simply
xfail the testcase for darwin8.
I put 10.5.3 up on my ppc64 platform last night and I have some
information for you:
1/ The test footprint (memory wise) is hugely improved on darwin9
[all gcc checks - not just libstdc++ ]
... which might point to some leakage in the toolchain used on
darwin8.
=====
2/ there is something not quite right with some tests:
the following tests cause "insane" malloc requests -m64 (which are
flagged on darwin9 ... but not on darwin8)
PASS: 21_strings/basic_string/cons/char/1.cc execution test
PASS: 21_strings/basic_string/cons/wchar_t/1.cc execution test
PASS: 21_strings/basic_string/insert/char/1.cc execution test
PASS: 21_strings/basic_string/insert/wchar_t/1.cc execution test
PASS: 23_containers/vector/bool/modifiers/insert/31370.cc execution test
EXAMPLE syslog error: "1.exe(1937) malloc: *** mmap
(size=4611686018427392000) failed (error code=12)
*** error: can't allocate region
*** set a breakpoint in malloc_error_break to debug"
NOTE: that all the tests are flagged as passed!
-m32 these tests do not cause any such request.
I do not know whether this is the intended behaviour ... but I
suspect not
=========================
The secondary effect ( on darwin 8) is this:
on darwin8
-m32 malloc is limited to sizes accommodated by most people's
machines (and avail root partition space)
-m64 does not appear to limit malloc size requests - such that they
can exceed the resources on the machine (I'll file a RADAR with apple
about this).
... the insane requests that these tests generate will kill real
machines ;-)
=========================
Please check that this is ppc-specific before XFAILing it --- note
that the tests are flagged as PASSED!
thanks,
Iain