Bug#674541: ruby1.8: segfaults when compiled with gcc-4.7

2013-03-19 Thread Colin Watson
On Sun, May 27, 2012 at 11:11:10PM +0100, Steven Chamberlain wrote:
 I discovered other segfaults during 'test all'.  An small, easily
 reproducible testcase is test/io/nonblock/test_flush.rb, which is fixed
 by compiling with the -fno-tree-dce compiler option.

ruby-ffi's spec/ffi/function_spec.rb test also fails every time when
ruby1.8 was built using gcc-4.7 without -fno-tree-dce, but is fixed by
adding -fno-tree-dce when building ruby1.8.  Just in case that helps
anyone ...

-- 
Colin Watson   [cjwat...@ubuntu.com]


-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org



Bug#674541: ruby1.8: segfaults when compiled with gcc-4.7

2012-05-27 Thread Steven Chamberlain
Hi,

I agree that building ruby1.8 with gcc-4.6 instead of gcc-4.7 (4.7.0-7),
with no other changes, makes the segfaults go away in the small threads
testcase.

But it is also okay if built using gcc-4.7 with only the
-fno-optimize-sibling-calls option added.

This is on kfreebsd-i386 and 4-way SMP, but clearly it has affected
Linux arches too (and even other distros).

Regards,
-- 
Steven Chamberlain
ste...@pyro.eu.org



-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org



Bug#674541: ruby1.8: segfaults when compiled with gcc-4.7

2012-05-27 Thread Steven Chamberlain
On 27/05/12 20:19, Steven Chamberlain wrote:
 But it is also okay if built using gcc-4.7 with only the
 -fno-optimize-sibling-calls option added.

That only took care of the small threads testcase I gave (verified with
~2000 runs).  But there seem to be other, separate issues.

I discovered other segfaults during 'test all'.  An small, easily
reproducible testcase is test/io/nonblock/test_flush.rb, which is fixed
by compiling with the -fno-tree-dce compiler option.


I've now tried gcc-4.7 4.7.0-9 from sid;  I found that it fixes the
small threads testcase (no need for -fno-optimize-sibling-calls any
more);  but the test_flush.rb testcase still always segfaults unless
-fno-tree-dce is used.

Regards,
-- 
Steven Chamberlain
ste...@pyro.eu.org



-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org