[OE-core] autom4te segfault in Dumper.so when configuring intltool

[Andre McCurdy]

On Wed, Apr 15, 2015 at 3:08 AM, Martin Jansa <martin.jansa at gmail.com> wrote:
> On Tue, Apr 14, 2015 at 07:05:35PM -0700, Andre McCurdy wrote:
>> On Tue, Apr 14, 2015 at 4:39 PM, Martin Jansa <martin.jansa at gmail.com> wrote:
>> > Yes, I see it very often since this patch almost 3 years ago:
>> > http://lists.openembedded.org/pipermail/openembedded-core/2012-August/066282.html
>>
>> The problem seems to be that autom4te is getting executed by the host
>> system's perl but loads Dumper.so from OE sysroot. If the host and OE
>> versions of perl are sufficiently different then host perl crashes
>> when it tries to execute OE sysroot Dumper.so
>>
>> Manually hacking the first line of autom4te to replace "/usr/bin/perl"
>> with the full path to the perl-native binary in OE sysroot makes the
>> crashes stop.
>
> Based on RP's reply in the old thread I assume you should rather try to
> hack it to load Dumper.so from system, so that there is no dependency
> between automake-native and perl-native.

Hiding OE sysroot perl when configuring intltool and using only host
perl does also prevent the segfault.

I'm not sure what the correct fix is though. Perhaps have the
autoconf-native scripts run host perl via a wrapper which removes any
reference to OE sysroot perl from the PATH?

>> > On Tue, Apr 14, 2015 at 11:34 PM, Andre McCurdy <armccurdy at gmail.com> wrote:
>> >>
>> >> I'm seeing messages such as the following appear in syslog when
>> >> building intltool or intltool-native from fido:
>> >>
>> >>   [1049484.121948] autom4te[16180]: segfault at 1 ip 00007fea41a7e7fe
>> >> sp 00007fff3c666080 error 4 in Dumper.so[7fea41a77000+8000]
>> >>   [1049567.199028] autom4te[17012]: segfault at 1 ip 00007f34acc007fe
>> >> sp 00007fff5ed84db0 error 4 in Dumper.so[7f34acbf9000+8000]
>> >>
>> >> The segfault happens reproducibly when configuring intltool or
>> >> intltool-native (one segfault every time I run or re-run the
>> >> run.do_configure script).
>> >>
>> >> There are no obvious errors in config.log and the intltool build
>> >> succeeds over-all.
>> >>
>> >> If I hack the intltool run.do_configure script so that the host
>> >> machine's perl (v5.18.2) is found instead of perl-native from sysroot
>> >> then the segfaults stop.
>> >>
>> >> Is anyone else seeing the same issue?
>> >>
>> >> Andre
>> >> --