[Debian-med-packaging] Another hdf5 issue [was: libminc has build test errors after using debhelper 10]

[Gilles Filippini]

Hi,

Andreas Tille a écrit le 24/11/2017 à 17:34 :
> Hi Gilles,
> 
> On Fri, Nov 24, 2017 at 12:43:46AM +0100, Gilles Filippini wrote:
>> Gilles Filippini a écrit le 23/11/2017 à 22:06 :
>>> Hi Andreas,
>>>
>>> These failures seem due mainly to the size change of type hid_t coming
>>> with HDF5 1.10.0.
>>> What is surprising is that nobody have encountered the issue before.
>>> It's almost one full year since I first uploaded HDF5 1.10.0 to unstable
>>> :/ Does this library have users actually?
> 
> I have no idea - I was just migrating this from SVN to Git and I'm
> listed as Uploaders only since I once touched the package and I tend to
> add myself as team uploader.
> 
>>> BTW it seems discontinued upstream: the last commit was in 2015 [1].
>>>
>>> [1] https://github.com/BIC-MNI/libminc
>>>
>>> I'll try to setup a patch, but I'm not sure I'll succeed, nor it's worth
>>> the amount of work.

No, it's not discontinued upstream actually. The work takes place into
the 'develop' branch [2].

[2] https://github.com/BIC-MNI/libminc/tree/develop

>> After a bit of digging, I've found this piece of code
>> (libsrc/hdf_convenience.c):
>>
>> MNCAPI int micreatex(const char *path, int cmode, struct mi2opts *opts_ptr)
>> {
>>     int fd;
>>
>>     MI_SAVE_ROUTINE_NAME("micreate");
>>
>>     if ((cmode & MI2_CREATE_V1) != 0) {
>>         fd = nccreate(path, cmode);
>>     }
>>     else if (miget_cfg_bool(MICFG_FORCE_V2) || (cmode & MI2_CREATE_V2)
>> != 0) {
>>         fd = hdf_create(path, cmode, opts_ptr);
>>     }
>>     else {
>>         if (mi_nc_files == 0 && mi_h5_files != 0) {
>>             /* Create an HDF5 file. */
>>             fd = hdf_create(path, cmode, opts_ptr);
>>         }
>>         else {
>>             /* Create a NetCDF file. */
>>             fd = nccreate(path, cmode);
>>         }
>>     }
>>     if (fd < 0) {
>>         milog_message(MI_MSG_CREATEFILE, path);
>>     }
>>     else {
>>         char ident[128];
>>
>>         micreate_ident(ident, sizeof(ident));
>>         miattputstr(fd, NC_GLOBAL, "ident", ident);
>>         miattputstr(fd, NC_GLOBAL, "minc_version", MINC_VERSION);
>>     }
>>     MI_RETURN(fd);
>> }
>>
>> When MICFG_FORCE_V2 or MI2_CREATE_V2 are set, the new file is created
>> using hdf_create(), which should return an HDF5 file handle of type hid_t.
>> But then this HDF5 file handle is passed to miattputstr() as a NetCDF
>> file handle o_O. Weird. I fail to see a happy end here.
>>
>> Any clue?
> 
> I do not have the slightest clue.  Steve?

This is fixed in upstream develop branch and was backported via an NMU
[3] to support HDF5 1.10.0. I just didn't remember that.

[3] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=841970

The actual problem is that this NMU has never been acknowledged in the
SVN repository.

Another problem is that the test suite apparently doesn't support being
run in parallel. You have to force non parallel mode with:

override_dh_auto_test:
	dh_auto_test --no-parallel

With the MNU changes and non parallel mode for dh_auto_test I've build
the library and run the test suite successfully.

Thanks,

_g.

P.S.: please check all the other packages you're migrating from SVN to
git for may be missing NMUs from the SVN repo.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 508 bytes
Desc: OpenPGP digital signature
URL: <http://lists.alioth.debian.org/pipermail/debian-med-packaging/attachments/20171126/00b42470/attachment.sig>