Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Gilles Filippini]

Julien Cristau a écrit le 03/09/2015 07:26 :
> On Thu, Sep  3, 2015 at 01:26:35 +0200, Gilles Filippini wrote:
> 
>> So where does this 'MUST NOT' come from just now that all that work is over?
>>
> It comes from me wanting to get gcc-defaults migrated this week.  Which
> is incompatible with getting all hdf5 reverse depends rebuilt.  It would
> probably have been fine a month ago when this started, but not now.

One month ago was family vacation time. Anyway, if the gcc-default
transition is to take place this week I guess the libhdf5-10 transition
can wait.

Consider that hdf5 1.8.13 in unstable is GCC-5 ready since it has no
more C++ rdepends left in testing after your removal of fw4spl and
insighttoolkit4 yesterday.

Thanks,

_g.



signature.asc
Description: OpenPGP digital signature

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Julien Cristau]

On Thu, Sep  3, 2015 at 01:26:35 +0200, Gilles Filippini wrote:

> So where does this 'MUST NOT' come from just now that all that work is over?
> 
It comes from me wanting to get gcc-defaults migrated this week.  Which
is incompatible with getting all hdf5 reverse depends rebuilt.  It would
probably have been fine a month ago when this started, but not now.

Cheers,
Julien


signature.asc
Description: Digital signature

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Gilles Filippini]

Julien Cristau a écrit le 02/09/2015 20:57 :
> Hrm.  Do I see correctly that hdf5 1.8.15 bumps SONAME of the C library?
> If so, that MUST NOT be uploaded to unstable before the C++ fun is over.
> If a transition is needed for the C++ lib, that must be done separately
> and ahead of the switch to 1.8.15.

You see correctly. And I see nothing contradictory with the directions
given by Matthias:
>  - If a library transition is needed, please prepare for the change.
>Rename the library package, append "v5" to the name of the package
>(e.g. libfoo2 -> libfoo2v5). *Such a change can be avoided, if you
>have a soversion bump and you upload this version instead of the
>renamed package*.

The soname bump was stated the 18th of july in this bug log [1] and I've
spend quite some hours checking that every rdepends builds fine. The
only few FTBFSs are either sid only, or low popcon with no rdepends, and
anyway not caused by hdf5 at all [2]

[1] 
[2] 

So where does this 'MUST NOT' come from just now that all that work is over?

Thanks,

_g.



signature.asc
Description: OpenPGP digital signature

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Julien Cristau]

On Wed, Sep  2, 2015 at 14:49:18 +0200, Sebastiaan Couwenberg wrote:

> On 02-09-15 14:37, Gilles Filippini wrote:
> > Gilles Filippini a écrit le 28/08/2015 00:12 :
> >> Sebastiaan Couwenberg a écrit le 27/08/2015 10:17 :
> >>> I suggest to reassign the hdf5 transition issue back to 
> >>> release.debian.org, because I would also like to see a single
> >>> hdf5 transition instead of two (I made the same decisions for
> >>> netcdf & geos).
> >> 
> >> Yes, I'll do that *after* having the rdeps checked to avoid
> >> useless ping pong game. I've started tracking progress on
> >> titanpad[1] in case anybody wants to help.
> >> 
> >> [1] 
> > 
> > I've tested a rebuild of all 68 rdepends reported on the auto-hdf5 
> > transition page [1]. Only 5 of them aren't binnmu OK:
> > 
> > [...]
> > 
> > To me, HDF5 1.8.15 is now transition ready.
> 
> Thanks for your work preparing the hdf5 transition, I'd say upload to
> unstable as soon as possible. We should have done hdf5 before netcdf.
> 
> The libvigraimpex transition (#794736) can then start triggered by the
> hdf5 transition.
> 
Hrm.  Do I see correctly that hdf5 1.8.15 bumps SONAME of the C library?
If so, that MUST NOT be uploaded to unstable before the C++ fun is over.
If a transition is needed for the C++ lib, that must be done separately
and ahead of the switch to 1.8.15.

Cheers,
Julien


signature.asc
Description: Digital signature

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Gilles Filippini]

Gilles Filippini a écrit le 02/09/2015 18:14 :
> Julien Cristau a écrit le 02/09/2015 17:56 :
>> On Wed, Sep  2, 2015 at 14:49:18 +0200, Sebastiaan Couwenberg wrote:
>>
>>> On 02-09-15 14:37, Gilles Filippini wrote:
 Gilles Filippini a écrit le 28/08/2015 00:12 :
> Sebastiaan Couwenberg a écrit le 27/08/2015 10:17 :
>> I suggest to reassign the hdf5 transition issue back to 
>> release.debian.org, because I would also like to see a single
>> hdf5 transition instead of two (I made the same decisions for
>> netcdf & geos).
>
> Yes, I'll do that *after* having the rdeps checked to avoid
> useless ping pong game. I've started tracking progress on
> titanpad[1] in case anybody wants to help.
>
> [1] 

 I've tested a rebuild of all 68 rdepends reported on the auto-hdf5 
 transition page [1]. Only 5 of them aren't binnmu OK:

 [...]

 To me, HDF5 1.8.15 is now transition ready.
>>>
>>> Thanks for your work preparing the hdf5 transition, I'd say upload to
>>> unstable as soon as possible. We should have done hdf5 before netcdf.
>>>
>> Does anything use the c++ hdf5 lib in Debian nowadays?  Last cycle it
>> didn't have any rdeps.
> 
> AFAICS, there is only fw4spl and libinsighttoolkit4.7.

and blasr (I previously only checked in testing).

Thanks,

_g.




signature.asc
Description: OpenPGP digital signature

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Gilles Filippini]

Julien Cristau a écrit le 02/09/2015 17:56 :
> On Wed, Sep  2, 2015 at 14:49:18 +0200, Sebastiaan Couwenberg wrote:
> 
>> On 02-09-15 14:37, Gilles Filippini wrote:
>>> Gilles Filippini a écrit le 28/08/2015 00:12 :
 Sebastiaan Couwenberg a écrit le 27/08/2015 10:17 :
> I suggest to reassign the hdf5 transition issue back to 
> release.debian.org, because I would also like to see a single
> hdf5 transition instead of two (I made the same decisions for
> netcdf & geos).

 Yes, I'll do that *after* having the rdeps checked to avoid
 useless ping pong game. I've started tracking progress on
 titanpad[1] in case anybody wants to help.

 [1] 
>>>
>>> I've tested a rebuild of all 68 rdepends reported on the auto-hdf5 
>>> transition page [1]. Only 5 of them aren't binnmu OK:
>>>
>>> [...]
>>>
>>> To me, HDF5 1.8.15 is now transition ready.
>>
>> Thanks for your work preparing the hdf5 transition, I'd say upload to
>> unstable as soon as possible. We should have done hdf5 before netcdf.
>>
> Does anything use the c++ hdf5 lib in Debian nowadays?  Last cycle it
> didn't have any rdeps.

AFAICS, there is only fw4spl and libinsighttoolkit4.7.

Thanks,

_g.




signature.asc
Description: OpenPGP digital signature

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Julien Cristau]

On Wed, Sep  2, 2015 at 14:49:18 +0200, Sebastiaan Couwenberg wrote:

> On 02-09-15 14:37, Gilles Filippini wrote:
> > Gilles Filippini a écrit le 28/08/2015 00:12 :
> >> Sebastiaan Couwenberg a écrit le 27/08/2015 10:17 :
> >>> I suggest to reassign the hdf5 transition issue back to 
> >>> release.debian.org, because I would also like to see a single
> >>> hdf5 transition instead of two (I made the same decisions for
> >>> netcdf & geos).
> >> 
> >> Yes, I'll do that *after* having the rdeps checked to avoid
> >> useless ping pong game. I've started tracking progress on
> >> titanpad[1] in case anybody wants to help.
> >> 
> >> [1] 
> > 
> > I've tested a rebuild of all 68 rdepends reported on the auto-hdf5 
> > transition page [1]. Only 5 of them aren't binnmu OK:
> > 
> > [...]
> > 
> > To me, HDF5 1.8.15 is now transition ready.
> 
> Thanks for your work preparing the hdf5 transition, I'd say upload to
> unstable as soon as possible. We should have done hdf5 before netcdf.
> 
Does anything use the c++ hdf5 lib in Debian nowadays?  Last cycle it
didn't have any rdeps.

Cheers,
Julien


signature.asc
Description: Digital signature

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Sebastiaan Couwenberg]

-BEGIN PGP SIGNED MESSAGE-
Hash: SHA512

On 02-09-15 14:37, Gilles Filippini wrote:
> Gilles Filippini a écrit le 28/08/2015 00:12 :
>> Sebastiaan Couwenberg a écrit le 27/08/2015 10:17 :
>>> I suggest to reassign the hdf5 transition issue back to 
>>> release.debian.org, because I would also like to see a single
>>> hdf5 transition instead of two (I made the same decisions for
>>> netcdf & geos).
>> 
>> Yes, I'll do that *after* having the rdeps checked to avoid
>> useless ping pong game. I've started tracking progress on
>> titanpad[1] in case anybody wants to help.
>> 
>> [1] 
> 
> I've tested a rebuild of all 68 rdepends reported on the auto-hdf5 
> transition page [1]. Only 5 of them aren't binnmu OK:
> 
> [...]
> 
> To me, HDF5 1.8.15 is now transition ready.

Thanks for your work preparing the hdf5 transition, I'd say upload to
unstable as soon as possible. We should have done hdf5 before netcdf.

The libvigraimpex transition (#794736) can then start triggered by the
hdf5 transition.

Kind Regards,

Bas

- -- 
 GPG Key ID: 4096R/6750F10AE88D4AF1
Fingerprint: 8182 DE41 7056 408D 6146  50D1 6750 F10A E88D 4AF1
-BEGIN PGP SIGNATURE-
Version: GnuPG v2
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=LACr
-END PGP SIGNATURE-

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Gilles Filippini]

Control: reassign -1 release.debian.org

Gilles Filippini a écrit le 28/08/2015 00:12 :
> Sebastiaan Couwenberg a écrit le 27/08/2015 10:17 :
>> I suggest to reassign the hdf5 transition issue back to
>> release.debian.org, because I would also like to see a single hdf5
>> transition instead of two (I made the same decisions for netcdf & geos).
> 
> Yes, I'll do that *after* having the rdeps checked to avoid useless ping
> pong game. I've started tracking progress on titanpad[1] in case anybody
> wants to help.
> 
> [1] 

Hi,

I've tested a rebuild of all 68 rdepends reported on the auto-hdf5
transition page [1]. Only 5 of them aren't binnmu OK:

fw4spl:
#797475, #797481 - FTBFS unrelated to hdf5
sid only

mapsembler2:
#797526 - FTBFS with GCC-5
Low popcon - no rdepends

gnudatalanguage:
waiting for plplot - #789619
sid only

feel++:
#777848 - FTBFS with GCC-5
sid only

insighttoolkit4:
#797387 - FTBFS unrelated to HDF5
marked for autoremoval from testing on 2015-09-06

To me, HDF5 1.8.15 is now transition ready.

Thanks,

_g.



signature.asc
Description: OpenPGP digital signature

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Gilles Filippini]

Sebastiaan Couwenberg a écrit le 27/08/2015 10:17 :
> I suggest to reassign the hdf5 transition issue back to
> release.debian.org, because I would also like to see a single hdf5
> transition instead of two (I made the same decisions for netcdf & geos).

Yes, I'll do that *after* having the rdeps checked to avoid useless ping
pong game. I've started tracking progress on titanpad[1] in case anybody
wants to help.

[1] 

Thanks,

_g.

P.S.:  I'll be on vac for the w-e.



signature.asc
Description: OpenPGP digital signature

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Sebastiaan Couwenberg]

-BEGIN PGP SIGNED MESSAGE-
Hash: SHA512

On 27-08-15 01:58, Gilles Filippini wrote:
> I just don't understand why this bug was assigned back to src:hdf5.
> A piece of explanation from vorlon would have been appreciated.
> Anyway, I'll assign it back to release.d.o when I'll have the
> rdepends builds checked.

I suspect this was because he's doing the GCC 5 transition in Ubuntu,
and they've chosen to do a v5 rename for hdf5 (like they've done for
netcdf and others too). He probably didn't want to do another
transition, but this is all an educated guess.

I suggest to reassign the hdf5 transition issue back to
release.debian.org, because I would also like to see a single hdf5
transition instead of two (I made the same decisions for netcdf & geos).

Kind Regards,

Bas

- -- 
 GPG Key ID: 4096R/6750F10AE88D4AF1
Fingerprint: 8182 DE41 7056 408D 6146  50D1 6750 F10A E88D 4AF1
-BEGIN PGP SIGNATURE-
Version: GnuPG v2
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=48Hu
-END PGP SIGNATURE-

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Gilles Filippini]

Hi Simon,

Thanks for helping :)

Simon McVittie a écrit le 27/08/2015 00:36 :
> On Sat, 18 Jul 2015 at 12:15:10 +0200, Gilles Filippini wrote:
>> hdf5 in experimental have a soversion bump. From there I'm unsure about
>> what to do right now.
> 
> This SONAME bump means the necessary rename has taken place in experimental.
> What do you plan to do about unstable?
> 
> There are basically two options:
> 
> * send the experimental version to unstable, if you believe this to
>   have a low risk of derailing the g++-5 transition, and in
>   particular all the reverse-dependencies still build;
> 
> * do a "v5" rename in unstable, similar to
>   https://patches.ubuntu.com/h/hdf5/hdf5_1.8.13+docs-15ubuntu2.patch
> 
> My advice would be to choose whichever of those you think is
> lower-risk.

IMHO either way is equal in risk. There are 60+ rdepends on hdf5 and I'd
very much prefer managing one direct transition to release 1.8.15 than
two transitions (to v5 then to 1.8.15).

> The release team have said[1] that starting the transition is OK for any
> library whose library dependencies have all either started *their*
> transitions, or been confirmed not to need transitions. hdf5
> doesn't appear to depend on any C++ libraries except mpich, which
> doesn't seem to be affected by this whole mess due to having a
> simpler ABI than most C++ libraries - so I think hdf5 is ready
> to go?

The hdf5 c++ lib is not built for any of the MPI variants (because not
supported upstream). Then, from what you wrote, I'd said that hdf5 is
ready to go.

But idealy I guess I'm expected to test the build of every rdepends,
which will take some time.

> (If you require confirmation from the release team I can try to
> get that, but I don't think they are going to be able to follow up
> on each of the dozens of parallel transitions involved in the libstdc++
> ABI change, which is why I'm trying to help keep things moving.)

I just don't understand why this bug was assigned back to src:hdf5. A
piece of explanation from vorlon would have been appreciated. Anyway,
I'll assign it back to release.d.o when I'll have the rdepends builds
checked.

Thanks again for your help!

_g.




signature.asc
Description: OpenPGP digital signature

Bug#791067: hdf5: library transition may be needed when GCC 5 is the default

[Simon McVittie]

On Sat, 18 Jul 2015 at 12:15:10 +0200, Gilles Filippini wrote:
> hdf5 in experimental have a soversion bump. From there I'm unsure about
> what to do right now.

This SONAME bump means the necessary rename has taken place in experimental.
What do you plan to do about unstable?

There are basically two options:

* send the experimental version to unstable, if you believe this to
  have a low risk of derailing the g++-5 transition, and in
  particular all the reverse-dependencies still build;

* do a "v5" rename in unstable, similar to
  https://patches.ubuntu.com/h/hdf5/hdf5_1.8.13+docs-15ubuntu2.patch

My advice would be to choose whichever of those you think is
lower-risk.

The release team have said[1] that starting the transition is OK for any
library whose library dependencies have all either started *their*
transitions, or been confirmed not to need transitions. hdf5
doesn't appear to depend on any C++ libraries except mpich, which
doesn't seem to be affected by this whole mess due to having a
simpler ABI than most C++ libraries - so I think hdf5 is ready
to go?

(If you require confirmation from the release team I can try to
get that, but I don't think they are going to be able to follow up
on each of the dozens of parallel transitions involved in the libstdc++
ABI change, which is why I'm trying to help keep things moving.)

Regards,
S

[1]