[Debian-med-packaging] Bug#982384: [Help] Bug 982384: Warnings profile count data file not found

[Helmut Grohne]

Hi Andreas,

On Mon, Dec 05, 2022 at 04:20:41PM +0100, Andreas Tille wrote:
> Hi Helmut,
> 
> Am Mon, Dec 05, 2022 at 05:25:03PM +0530 schrieb Nilesh Patra:
> > 
> > On 5 December 2022 12:24:23 pm IST, Andreas Tille <tille at debian.org> wrote:
> > >Since git blames you about last changes on the lines with the option in
> > >question could you please comment about this?

The assumption that the one who last touched this is at fault is an
interesting one. Unfortunately, it is quite distant from the truth.

> > I simply applied patch supplied by Helmut to make build cross-buildable. Please check Bug#989942 for the explanation and context.

The next step is forwarding the question to the next person until
someone is found who does actual research. :-/

> Bug #982384 (libdeflate: Warnings profile count data file not found)
> implies that the patch you provided once is not working as expected.
> Would you mind commenting on this issue?

I think the matter is relatively simple. You say that the
-Wmissing-profile warnings shouldn't be there and I agree. You also
imply that they haven't been there forever. So we're likely talking
about a regression here, right?

Once we assume to have a regression, there is a dead simple way of
figuring out: Just go back in the build logs and locate a version that
doesn't contain these warning. I've done this for your convenience and
it turns out that 1.2-1 is the last version that lacks these warnings.
1.3-1 is the first that contains them. The changelog for 1.3-1 is:

| libdeflate (1.3-1) unstable; urgency=medium
|
|   * New upstream version, now provides libdeflate-g{un,}zip in /usr/bin
|   * debhelper-compat 12
|   * Standards-Version: 4.4.0
|   * Remove trailing whitespace in debian/rules
|   * Set upstream metadata fields: Name.
|
|  -- Michael R. Crusoe <michael.crusoe at gmail.com>  Fri, 23 Aug 2019 12:28:07 +0200

I think this would be a better start for research than blaming random
patches.

While we can now start blaming someone else, we should not rule out
possible toolchain changes as a cause given that 1.2-1 and 1.3-1 are
half a year apart and were built with gcc 8.2 and gcc 9.2 respectively.
Quite simply, it could be that the profile guided optimization never
worked, but only gcc 9 would start warning about it. Maybe we could
check when that warning was added to gcc?

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86957

That happens to be gcc 9. Surprise!

So while we don't have proof yet, it seems quite likely that profile
guided optimization never worked for libdelfate. Just now, gcc tells us
about that.

Hope this helps

Helmut