Re: Bug in apt 0.5.15lorg2 using rpm 4.4.5

On Sun, 2006-03-05 at 18:42 -0300, Herton Krzesinski wrote:
> Hi, I saw that Panu is continuing work for apt-rpm, In using his new
> apt a friend (Andreas Hasenack) found a bug in apt-get build-dep,
> basically an apt-get build-dep <some file>.src.rpm doesn't work, it
> fail on rpm internal dependencies. Looking trough the code I found the
> problem and made a patch, I'm attaching it here. Here is the
> description of internals about the patch:

Ah.. indeed. Thanks for spotting this :)

> There is an error inside new apt-rpm 0.5.15lorg2 inside the function 
> rpmSrcRecordParser::BuildDepends at file rpmsrcrecords.cc, as the
> function was failing to detect/pass the internal rpm dependencies.
> Patch details: If you want to check the results after rpmdsSearch(*,
> ds) with rpmdsResult, before Search you must use the function
> rpmdsSetResult (eg.: rpmdsSetResult(ds, 0)) to alloc memory  for
> ds->Result, otherwise rpmdsSearch will not be able to store
> information in ds->Result.
> 
> The fix could be different of course. We could avoid to use
> rpmdsResult, just checking what rpmdsSearch returned (a value >=0 the
> key was found, like is used just lines above with getconf for
> example). Or if you consider the problem of ds->Result not being
> allocated a bug of rpm, then rpmdsSingle must alloc memory for it. I
> don't know if there is a right thing to do here or if the problem is
> really of apt, but I decided to patch apt-rpm and this fixes the issue
> of build-dep not working.

I don't actually know how that piece of code even got there, the same
thing is in rpmlistparser.cc in a working format.. I've fixed that for
now, the "real fix" for this stuff is to stuff the dependency parsing
into a single place to avoid copy-paste errors such as this, as the
dependency information is handled just the same for both binary and
source rpms.

	- Panu -