svn commit: r521506 - head/science/hdf5

[Piotr Kubaj]

On 20-01-01 10:56:30, Jason Bacon wrote:
> On 2020-01-01 09:30, Jason Bacon wrote:
> > On 2020-01-01 03:46, Yuri wrote:
> > > On 2019-12-30 08:57, Sunpoet Po-Chuan Hsieh wrote:
> > > > Log:
> > > >    Update to 1.10.6
> > > >       Changes: https://portal.hdfgroup.org/display/support/HDF5+1.10.6
> > > > 
> > > > Modified:
> > > >    head/science/hdf5/Makefile
> > > >    head/science/hdf5/distinfo
> > > >    head/science/hdf5/pkg-plist
> > > 
> > > 
> > > This commit broke science/rmf
> > > 
> > > 
> > > Yuri
> > > 
> > Perhaps we could develop a poudriere option to test all dependents as
> > well as dependencies?  Obviously this would increase testing time
> > significantly for some ports, but making thorough testing convenient for
> > the developer is the main thing.
> > 
> > An easy way to prevent breakages like this one would be a great
> > improvement to the ports system.  Even an imperfect solution would be a
> > huge help here, I think.
> > 
> >     JB
> > 
> Or just script around it.  The hack below should list virtually all direct
> dependent ports, though there might be weird edge cases it misses.
> 
> #!/bin/sh
> 
> usage()
> {
>     printf "Usage: $0 category/port\n"
>     exit 1
> }
> 
> if [ $# != 1 ]; then
>     usage
> fi
> 
> : ${PORTSDIR:=/usr/ports}
> cd $PORTSDIR
> 
> # One category at a time to avoid "argument list too long"
> # Combine '\' continued *_DEPENDS lines and check for category/port.
> for category in *; do
>     if [ $category != Templates ] && [ -e $category/Makefile ]; then
>     for port in $category/*; do
>         if [ -e $port/Makefile ] && \
>             awk '{ if (sub(/\\$/,"")) printf "%s", $0; else print $0 }' \
>             $port/Makefile* | grep -q "_DEPENDS.*:$1"; then
>         echo $port
>         fi
>     done
>     fi
> done
> 
> hdf5 has some big dependents (e.g. octave) so it would take a while to test
> them all, but I'm not sure there's any other way to be certain.  Most ports
> don't have many big dependents, so it wouldn't cost much to test them all. 
> Also, testing dependents on one OS version and architecture would probably
> catch almost all the issues.

I disagree, there are many issues that happen only on non-x86 architectures.
The usual causes are:
- different base compiler,
- lack of SSE,
- different endianness,
- unsigned vs signer char.

Simple ports will build everywhere, but more sophisticated ones don't have to.

> 
> FreeBSD coral.acadix  bacon ~ 790: dependent-ports science/hdf5
> audio/csound
> biology/kallisto
> cad/freecad
> cad/gmsh
> devel/py-tables
> french/aster
> french/med
> graphics/alembic
> graphics/blender
> graphics/caffe
> graphics/gdal
> graphics/opencv
> graphics/openimageio
> graphics/qgis
> graphics/qgis-ltr
> graphics/vigra
> graphics/vips
> graphics/vv
> math/ambit
> math/armadillo
> math/deal.ii
> math/dynare
> math/flann
> math/labplot
> math/libmesh
> math/mathgl
> math/matio
> math/moab
> math/octave
> math/pdal
> math/scilab
> science/ALPSCore
> science/avogadro2
> science/avogadrolibs
> science/cdo
> science/cgnslib
> science/chemps2
> science/chrono
> science/code_saturne
> science/erkale
> science/fleur
> science/gnudatalanguage
> science/h5utils
> science/h5z-zfp
> science/highfive
> science/InsightToolkit
> science/jdftx
> science/kst2
> science/lm
> science/meep
> science/minc2
> science/mpb
> science/netcdf
> science/paraview
> science/py-h5py
> science/py-netCDF4
> science/qmcpack
> science/rmf
> science/silo
> sysutils/slurm-wlm
> 
> -- 
> Earth is a beta site.
> 
> 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/svn-ports-head/attachments/20200101/1b64c845/attachment.sig>