Importance of debuginfo packages

[Jason L Tibbitts III]

Often when doing package reviews I find a package that needs some hack
in order to get rpmbuild to build a complete debuginfo package.  Some
of this relates to passing the proper compiler flags and making sure
the package doesn't strip its own binaries, which I think is
necessary, but occasionally rpmbuild needs the source to be moved
around in the build tree, or it needs symlinks flattened or somesuch.

These are just hacks to work around rpmbuild deficiencies, and I can
understand if packagers are reticent to clutter up their packages for
something of unknown value.  So, how important is it to have a proper
debuginfo package?  How much hacking is permissible or necessary in
order to achieve that goal?  Is an incomplete debuginfo package (that
contains, say, the striped debug information but is missing some or
all of the source code) still useful, or would it be better to have
none at all?

- J<

[Rex Dieter]

Jason L Tibbitts III wrote:
> So, how important is it to have a proper
> debuginfo package?  How much hacking is permissible or necessary in
> order to achieve that goal?

I think these are questions best answered by the maintainer/reviewer 
(and possibly upstream).  The easy answers are:
tiny hacks -> do it
many hacks/patches -> probably not worth it
It's defining the line between the two that's tricky.
> Is an incomplete debuginfo package (that
> contains, say, the striped debug information but is missing some or
> all of the source code) still useful, or would it be better to have
> none at all?

Something is certainly better than nothing.

-- Rex

[Jason L Tibbitts III]

> > > > > "RD" == Rex Dieter  writes:

RD> Something is certainly better than nothing.

Is it really?  I don't know how to use debuginfo packages, so I don't
know if there's any point to having them if they're not complete.

Obviously we know that a debuginfo package with no files should be
disabled, so what needs to be there before there's any point to
enabling it?

- J<

[Jesse Keating]

On Monday 21 August 2006 12:26, Jason L Tibbitts III wrote:
> Is it really?  I don't know how to use debuginfo packages, so I don't
> know if there's any point to having them if they're not complete.
>
> Obviously we know that a debuginfo package with no files should be
> disabled, so what needs to be there before there's any point to
> enabling it?

Enough so that running gdb (gnu debugger) on the application will produce a 
meaningful traceback.  This helps a lot in figuring out where / why an 
application is crashing.  Without a debuginfo package you cannot get this 
information without recompiling the package and leaving the binaries 
unstripped.

[Jason L Tibbitts III]

> > > > > "JK" == Jesse Keating  writes:

JK> Enough so that running gdb (gnu debugger) on the application will
JK> produce a meaningful traceback.

So what needs to be there in order for that to work?

I need to be able to superficially inspect a debuginfo package and
know whether it has enough stuff in it or not.

- J<

[Axel Thimm]

On Mon, Aug 21, 2006 at 11:54:22AM -0500, Jason L Tibbitts III wrote:
> > > > > > "JK" == Jesse Keating  writes:
>
> JK> Enough so that running gdb (gnu debugger) on the application will
> JK> produce a meaningful traceback.
>
> So what needs to be there in order for that to work?
>
> I need to be able to superficially inspect a debuginfo package and
> know whether it has enough stuff in it or not.

That's very difficult, especially in projects with subdirectories that
may redefine CFLAGS/CXXFLAGS w/o the builder noticing. sope/ogo is one
nasty candidate for example.

There are some safety pins in debuginfo, IIRC if it's turned on, but
the debuginfo turns out empty the build will fail. That's how it gets
my attention to fix debugging symbols at least :)

[Tom Callaway]

On Mon, 2006-08-21 at 19:16 +0200, Axel Thimm wrote:
> On Mon, Aug 21, 2006 at 11:54:22AM -0500, Jason L Tibbitts III wrote:
> > > > > > > "JK" == Jesse Keating  writes:
> >
> > JK> Enough so that running gdb (gnu debugger) on the application will
> > JK> produce a meaningful traceback.
> >
> > So what needs to be there in order for that to work?
> >
> > I need to be able to superficially inspect a debuginfo package and
> > know whether it has enough stuff in it or not.
>
> That's very difficult, especially in projects with subdirectories that
> may redefine CFLAGS/CXXFLAGS w/o the builder noticing. sope/ogo is one
> nasty candidate for example.
>
> There are some safety pins in debuginfo, IIRC if it's turned on, but
> the debuginfo turns out empty the build will fail. That's how it gets
> my attention to fix debugging symbols at least :)

Indeed, debuginfo being empty is about the only giant red warning flag
that I'm aware of. :)

~spot

[Parag Nemade]

Hi,
On 8/21/06, Tom 'spot' Callaway  wrote:
> On Mon, 2006-08-21 at 19:16 +0200, Axel Thimm wrote:
> > On Mon, Aug 21, 2006 at 11:54:22AM -0500, Jason L Tibbitts III wrote:
> > > > > > >> "JK" == Jesse Keating  writes:
> > >
> > > JK> Enough so that running gdb (gnu debugger) on the application will
> > > JK> produce a meaningful traceback.
> > >
> > > So what needs to be there in order for that to work?
> > >
> > > I need to be able to superficially inspect a debuginfo package and
> > > know whether it has enough stuff in it or not.
> >
> > That's very difficult, especially in projects with subdirectories that
> > may redefine CFLAGS/CXXFLAGS w/o the builder noticing. sope/ogo is one
> > nasty candidate for example.
> >
> > There are some safety pins in debuginfo, IIRC if it's turned on, but
> > the debuginfo turns out empty the build will fail. That's how it gets
> > my attention to fix debugging symbols at least :)
>
> Indeed, debuginfo being empty is about the only giant red warning flag
> that I'm aware of. :)

May i know if debuginfo packages are really good in terms of let the
end user debug applications to check if he is facing any crashes then
why not we have separate debuginfo directory also present publicly for
download?

Regards,
Parag.

[José Abílio Matos]

On Sunday 10 September 2006 09:51, Parag N(पराग़) wrote:
> May i know if debuginfo packages are really good in terms of let the
> end user debug applications to check if he is facing any crashes then
> why not we have separate debuginfo directory also present publicly for
> download?

But they are. :-)
  Look into /etc/yum.repos.d/fedora-extras.repo the repository is there 
although disabled by default.
> Regards,
> Parag.

[Parag Nemade]

Hi,
On 9/10/06, José Matos  wrote:
> On Sunday 10 September 2006 09:51, Parag N(पराग़) wrote:
> > May i know if debuginfo packages are really good in terms of let the
> > end user debug applications to check if he is facing any crashes then
> > why not we have separate debuginfo directory also present publicly for
> > download?
>
> But they are. :-)
>   Look into /etc/yum.repos.d/fedora-extras.repo the repository is there
> although disabled by default.

Oops i did not look into that. I should have checked it before. One
more thing there is no separate iso or directory present in install
media why so? Does that mean whoever want to use debuginfo packages he
must download it from fedora website?
> > Regards,
> > Parag.

[José Abílio Matos]

On Monday 11 September 2006 05:07, Parag N(पराग़) wrote:
> Oops i did not look into that. I should have checked it before. One
> more thing there is no separate iso or directory present in install
> media why so?

Due to disk space constraints and since those packages are seldom needed the 
compromise was to leave them only on the on-line repository.
> Does that mean whoever want to use debuginfo packages he 
> must download it from fedora website?

Yep.