[Pkg-db-devel] Bug#437709: marked as done (/usr/sbin/dhelp_parse: /usr/lib/libdb-4.5.so: no version information available (required by /usr/sbin/dhelp_parse))

[Florian Weimer]

* Clint Adams:

> On Sun, Oct 07, 2007 at 04:39:06PM +0000, Debian Bug Tracking System wrote:
>>    * Bump shlibs due to the introduction of symbol versioning.
>>      Closes: #437709.
>>    * Disable test suite on hppa.
>
> I don't think either of these changes was necessary.  With the test
> suite disabled, we will now not get any data confirming that the
> build hanging has ceased due to the mutex change, 

According to

<http://buildd.debian.org/fetch.cgi?&pkg=db4.5&ver=4.5.20-8&arch=hppa&stamp=1190420744&file=log>

the test suite has failed completely on hppa.  I see no point in running
it any further and risking that testing propagation is affected by hppa.

It seems to me that your mutex change on hppa somehow disabled shared
environments, so if you fix that, it might make sense to enable the test
suite again, at least for one upload.  (The lack of shared environment
support might well be RC. *sigh* [1])

> and since db4.5 was not in etch, the lack of symbol versioning in -1
> was not a bug, but merely a problem which would have disappeared with
> eventual propagation to testing.

I'm not completely sure that the buildds will pick up the new version
unless there is a proper dependency.  The issue would have gone away
eventually, when all the reverse dependencies have been rebuilt, but
there is no way to count on that.  At the very least, it's now apparent
from the Packages file if something has been built with the correct
version, so we can actually ask the release team to sort this out using
bin-NMUs.

[1] Turns out that the following two calls fail in the pthreads test
case:

pthread_condattr_setpshared(&condattr, PTHREAD_PROCESS_SHARED)
pthread_mutexattr_setpshared(&mutexattr, PTHREAD_PROCESS_SHARED)

IOW, mutexes may not cross process boundaries.  This is a libc bug, I
suppose.