Make empty debuginfo packages a build error?

[Ville Skyttä]

Hi,

How about making empty debuginfo packages a build error?  I can't think of a 
reason why an empty one would be a desired result, but empty ones do almost 
always indicate problems with the build.  Arch specific packages that 
expectedly don't have anything to place into the debuginfo package should 
explicitly disable it eg. with "%define debug_package %{nil}".

Also submitted as https://bugzilla.redhat.com/249956

[Jason L Tibbitts III]

> >>>> "VS" == Ville Skyttä
> <ville.skytta(a)iki.fi&gt; writes: 
VS> How about making empty debuginfo packages a build error?

Sure.  And I'm having trouble coming up with a situation where a
debuginfo containing a single file is OK, since it generally indicates
that debuginfo was extracted from an executable but no source was
found.  (Common for Java packages, at least.)  Or is such a stunted
debuginfo package still considered useful?

 - J<

[Ville Skyttä]

On Saturday 28 July 2007, Jason L Tibbitts III wrote:
>  >>>>> "VS" == Ville Skyttä
> <ville.skytta(a)iki.fi&gt; writes:
>
>  VS> How about making empty debuginfo packages a build error?
>
>  Sure.  And I'm having trouble coming up with a situation where a
>  debuginfo containing a single file is OK, since it generally indicates
>  that debuginfo was extracted from an executable but no source was
>  found. (Common for Java packages, at least.) 
Yep, ditto cases where "strip -g" is invoked on the binaries before 
find-debuginfo.sh does its thing, or the binaries are built without -g which 
almost always means that $RPM_OPT_FLAGS aren't being used either and at that 
point useless -debuginfo is no longer necessarily the biggest problem.

>  Or is such a stunted  
>  debuginfo package still considered useful? 
I don't think so, at least generally, but detecting it without false positives 
is harder than the empty debuginfo case.  And the Java case is a good example 
of another problem - I'm not sure we'd want people to disable debuginfo 
packages because the tools don't support creating sane ones from them yet.

[Jason L Tibbitts III]

> >>>> "VS" == Ville Skyttä
> <ville.skytta(a)iki.fi&gt; writes: 
VS> And the Java case is a good example of another problem - I'm not
VS> sure we'd want people to disable debuginfo packages because the
VS> tools don't support creating sane ones from them yet.

Actually it's possible to fix the Java packages; you have to move the
source so that the scripts will find it.  I've been through this on
pretty much all of the Java packages I reviewed.  (It's been a while
since I did one, though.)  I've also had to do this with some Perl
XS-built packages.

I admit this is nasty, but I've been assuming that we really want
proper debuginfo when we can get it.

 - J<

[Ville Skyttä]

On Saturday 28 July 2007, Jason L Tibbitts III wrote:
>  >>>>> "VS" == Ville Skyttä
> <ville.skytta(a)iki.fi&gt; writes:
>
>  VS> And the Java case is a good example of another problem - I'm not
>  VS> sure we'd want people to disable debuginfo packages because the
>  VS> tools don't support creating sane ones from them yet.
>
>  Actually it's possible to fix the Java packages; you have to move the
>  source so that the scripts will find it.  I've been through this on
>  pretty much all of the Java packages I reviewed.  (It's been a while
>  since I did one, though.)  I've also had to do this with some Perl
>  XS-built packages. 
Ah, good to know.  If there's a common pattern how these files need to be 
moved around or how to generally find a way to do it, could you add some info 
about it to http://fedoraproject.org/wiki/Packaging/Debuginfo ?

>  I admit this is nasty, but I've been assuming that we really
> want
>  proper debuginfo when we can get it. 
Ditto.

[Jason L Tibbitts III]

> >>>> "VS" == Ville Skyttä
> <ville.skytta(a)iki.fi&gt; writes: 
VS> If there's a common pattern how these files need to be moved
VS> around or how to generally find a way to do it, could you add some
VS> info about it to http://fedoraproject.org/wiki/Packaging/Debuginfo
VS> ?

I'm not sure that I even remember how to do it.  I guess I could
review another Java package, or I could try to find an on-point review.

Here's one:
  http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=191014

I note that you've commented in there as well.

 - J<

[Till Maas]

On Saturday 28 July 2007 19:08:03 Jason L Tibbitts III wrote:

>  Actually it's possible to fix the Java packages; you have to move
> the
>  source so that the scripts will find it.  I've been through this on 
Is it not possible to fix the script to find the source files without moving 
them? Or maybe add an enviroment variable to specify where it should search?

Regards,
Till

[Jason L Tibbitts III]

> >>>> "TM" == Till Maas
> <opensource(a)till.name&gt; writes: 
TM> Is it not possible to fix the script to find the source files
TM> without moving them?

Honestly I know nothing about how the debuginfo is generated and I'm
loath to poke around in there.

 - J<