On Sep 10, 2012, at 1:48 AM, Jan Rękorajski wrote:
> On Sun, 09 Sep 2012, Jeffrey Johnson wrote:
>
>>
>> On Sep 9, 2012, at 3:57 PM, Jan Rękorajski <bagg...@sith.mimuw.edu.pl> wrote:
>>
>>>
>>> Looks it's not a ldconfig optimization that's the problem, just that
>>> scriplets with only '-p /sbin/ldconfig' interpreter and empty body are
>>> not run.
>>> Function runInstScript() in lib/psm.c exits on 'headerGet(fi->, She, 0)' in
>>> such cases.
>>
>> Yes busted with %post -p /sbin/ldconfig (though the
>> cause may be not adding the script tag when empty
>> during build, entirely different root cause: will look).
>>
>>> I don't want to blindly poke in the code, so the question - would it be
>>> okay if that function bailed out only if headerGet failed on both
>>> psm->progTag and psm->scriptTag, instead of any of them?
>>>
>>
>> The code looks scarier than it actually is, largely because
>> of endless diddle-ups.
>>
>> When return code fails on non-existent script, malloc an empty string.
>>
>> So (untested)
>> if (!headerGet(fi->h, She, 0))
>> She->p.str = xstrdup("");
>>
>> See if that fixes %post -p /sbin/ldconfig for you.
>
> Close, but no cigar:
>
> /sbin/ldconfig: relative path `2' used to build cache
> error: %post(libevent-2.0.14-2.x86_64) scriptlet failed, exit status 1
>
These are $1 $2 arguments passed to /sbin/ldconfig by RPM, another sick hack
(in glibc) to permit
%post -p /sbin/ldconfig
> Putting empty script there must be doing harm because merging the
> conditions like this:
>
> if (!headerGet(fi->h, She, 0) && !headerGet(fi->h, Phe, 0))
> goto exit;
>
> fixes my problem nicely.
>
As you've seen, not quite.
There are likely other values than She->p.str that should be initialized that
are changing logic flow.
It would be nice to get a real fix, not just ignoring an error return, in place.
Something else is occurring ...
73 de Jeff
______________________________________________________________________
RPM Package Manager http://rpm5.org
Developer Communication List rpm-devel@rpm5.org
