tests of 1.7 fail on Solaris 11 x86

[GoogleCodeExporter]

4 failures, see attached check.log
Solaris 11 was run as a VirtualBox guest on FreeBSD 8 host.
Solaris 11 now has gcc-3.4.3 pre-installed as a default compiler. It was used 
to build perftools.

Original issue reported on code.google.com by yuriv...@yahoo.com on 10 Jun 2011 at 6:04

Attachments:

• check.log

[GoogleCodeExporter]

debugallocation_test may be the same as with issue 342: it just needs the 
bugfix present in svn-root.

heap-profiler_unittest is known to be flaky on freebsd (see the INSTALL file).  
I wonder if that flakiness is bleeding into the virtual box.  It doesn't seem 
like it though, since we're seeing errors like
   Hooked allocator frame not found, returning empty trace
and
   google_malloc section is missing, thus InHookCaller is broken!

I don't know if this is a solaris issue, gcc-3.4.3 issue, or something else.  I 
actually have a solaris-x86 machine to test on, and these tests pass for me 
there.  I think it uses a more modern gcc, though.

I see the same messages for the failed heap-checker-death_unittest.sh.  In 
fact, I'm surprised the heap-checker unittest passed.  It should have the same 
errors.  In any case, the heap-checker uses linux-specific code, and shouldn't 
work on solaris at all.  The configure script is supposed to notice you're 
running on solaris and not even configure or run the heap-checker test.  Do you 
know why it did?

Original comment by csilv...@gmail.com on 10 Jun 2011 at 6:45

• Added labels: Priority-Medium, Type-Defect

[GoogleCodeExporter]

I don;t think FreeBSD should matter for this case. I only mentioned it to 
describe how I run it, but once things are in VBox they pretty much run as if 
in guest OS which is Solaris 11.

I see from configure script that it sets the value for 
default_enable_heap_checker to "no" for mingw, cygwin, freebsd, darwin and to 
"yes" for everyone else, including solaris. It's this way in both trunk and 1.7.

Original comment by yuriv...@yahoo.com on 10 Jun 2011 at 8:25

[GoogleCodeExporter]

Blah, I've led you astray *again*.  I dusted off my solaris machine and managed 
to run the tests again, and I'm also getting failing tests.  (More than you are 
since I don't have objdump/nm installed either.)  I think my failures are a 
superset of yours.  So I'm changing this to a porting issue; I think solaris 
just isn't robust to support all of the advanced perftools tools yet.  The 
basic tcmalloc_minimal should work though.

Original comment by csilv...@gmail.com on 10 Jun 2011 at 9:07

• Added labels: Type-Porting
• Removed labels: Type-Defect

[GoogleCodeExporter]

So what is missing on Solaris? nm actually came pre-installed (/usr/gnu/bin/nm) 
on my Solaris 11 which I installed as a VBox appliance. Maybe the issue is that 
my newest binutils were superseded by this original binutils?

Solaris is heavily promoted by Oracle now on new DB servers. I think perftools 
better supports it.

Original comment by yuriv...@yahoo.com on 10 Jun 2011 at 9:17

[GoogleCodeExporter]

We'll happily support any OS we can! (within reason.)  Patches gladly accepted.

Original comment by csilv...@gmail.com on 10 Jun 2011 at 10:07

[GoogleCodeExporter]

I've decided to close all the porting requests that don't have patches attached 
to them.  We're always happy to look at patches for new architectures and OSes, 
but otherwise we are limited to the CPUs and OSes mentioned in the INSTALL file.

Original comment by csilv...@gmail.com on 31 Aug 2011 at 1:15

• Changed state: WontFix