Re: libffi 3.0.9 release candidate 1

[Andreas Tobler <andreast-list@fgznet.ch>]

Jasper Lievisse Adriaanse wrote:
> On Mon, Oct 05, 2009 at 12:57:30AM -0400, Anthony Green wrote:
>> Here's 3.0.9rc1:
>> ftp://sources.redhat.com/pub/libffi/libffi-3.0.9rc1.tar.gz
>>
>> I built this out of my new git repository here:
>> http://github.com/atgreen/libffi
>>
>> I'm using stgit, which is like quilt for git.  I rebased the git
>> repository to gcc HEAD, then added the "stand-alone" patches (mostly
>> configury changes), the AVR port, the Snow Leopard memory protection
>> fix, and the 64-bit AIX fix.
>>
>> I believe we're missing ChangeLog entries from Jens, Abdulaziz, and
>> Bradley.  Please send them.  I'll push your changes into GCC once
>> they're clean.
>>
>> Abdulaziz - please also check that I applied your patch correctly.
>> The source files had changed somewhat since you submitted it.
>>
>> I'll set up a wiki page to track platform testing soon, but initial
>> test results and pointers to missing patches would be helpful.
>>
>> Thanks,
>>
>> AG
> 
> As of this RC, there's an increase of about 340 in failed tests on
> OpenBSD. I've attached the log. The unwindtest.cc failure is expected,
> ther others not. Anyone has a clue what's going on here?
> 
> libffi 3.0.8:
> # of expected passes            1365
> # of unexpected failures        8
> # of unsupported tests          15
> 
> libffi 3.0.9rc1
> # of expected passes            685
> # of unexpected failures        348
> # of unsupported tests          15

Could you try to copy the testcases from 3.0.8 (maybe even 3.0.7) to 
3.0.9rc1 and rerun the testsuite?

I encounter similar failures on FreeBSD sparc64 and amd64. Copying the 
testcases from an older libffi release to the 3.0.9rc1 brings everthing 
back in PASS land here. (with gcc-4.2.x)
Well, that is not the solution, just a first step to isolate the issue.

In the meantime I try to find out what is going on, I guess MMAP vs. 
ffi_closure_alloc is the starting point to investigate.

Andreas