On 08/13/2017 06:24 PM, Orion Poplawski wrote:
On 08/13/2017 04:09 AM, Panu Matilainen wrote:
On 08/12/2017 06:29 PM, Orion Poplawski wrote:
On 08/10/2017 01:47 PM, Panu Matilainen wrote:
Resending to devel@, some mishap in earlier...
-------- Forwarded Message --------
Subject: Heads-up on rpm 4.14 coming to rawhide, incl. soname bump
Date: Thu, 10 Aug 2017 20:50:30 +0300
From: Panu Matilainen <pmatilai@xxxxxxxxxx>
To: devel-announce@xxxxxxxxxxxxxxxxxxxxxxx, Development discussions
related to Fedora <devel@xxxxxxxxxxxxxxxxxxxxxxx>
Rpm 4.14 alpha is about to hit rawhide as per
https://fedoraproject.org/wiki/Changes/RPM-4.14. I realize we're a
bit late to the schedule, and much much later in the cycle than I'd
like, but lets try to get by. There's a soname bump involved so
related packages will need to be rebuilt, but Igor Gnatenko kindly
promised to take care of that.
The other thing is the rpm macro engine which has seen by far its
biggest in the last 10+ years:
- Macro scoping is simplified, only macros %define'd inside
parametric macros are local, everything else is in global scope.
This isn't as big a change as is it may sound, because this is the
way scoping only ever really worked in previous versions.
- %{lua:} macros are scoped similarly to everything else. Previously
they could appear to be higher in the callchain than they are,
causing strange side-effects in already hard to debug complex macro
constructs.
- Visibility is enforced per scope for automatic macros (ie %* %1
etc that are created for parametric macros). In other words, nested
macro "calls" are more reliable as arguments from previous calls are
not visible.
- Arguments to parametric macros are now expanded before execution.
- Arguments to parametric macros now support quoting (single and
double)
For the average trivial macro usage, none of that will make a
difference. But I'd be almost shocked if nothing at all broke
because of this - former limitations and bugs might have necessiated
tricks and kludges that no longer work but also are not required any
more etc. So owners of complex macros, keep your eyes open.
Okay, so this has hit the octave macro set, probably not
unexpectedly. I'd appreciate some help with what I'm doing wrong, if
anything:
ENTER ['do'](['bash', '--login', '-c', '/usr/bin/rpmbuild -bb
--target noarch --nodeps
/builddir/build/SPECS/octave-statistics.spec'],
logger=<mockbuild.trace_decorator.getLog object at
0x3fff945826d8>timeout=172800user='mockbuild'printOutput=Falseenv={'SHELL':
'/bin/bash', 'HOME': '/builddir', 'PROMPT_COMMAND': 'printf
"\\033]0;<mock-chroot>\\007"', 'PATH':
'/usr/bin:/bin:/usr/sbin:/sbin', 'LANG': 'en_US.UTF-8', 'TERM':
'vt100', 'PS1': '<mock-chroot> \\s-\\v\\$ ', 'HOSTNAME':
'mock'}gid=425uid=1000shell=FalsechrootPath='/var/lib/mock/f27-build-9511079-776071/root'nspawn_args=[])
Executing command: ['bash', '--login', '-c', '/usr/bin/rpmbuild -bb
--target noarch --nodeps
/builddir/build/SPECS/octave-statistics.spec'] with env {'SHELL':
'/bin/bash', 'HOME': '/builddir', 'PROMPT_COMMAND': 'printf
"\\033]0;<mock-chroot>\\007"', 'PATH':
'/usr/bin:/bin:/usr/sbin:/sbin', 'LANG': 'en_US.UTF-8', 'TERM':
'vt100', 'PS1': '<mock-chroot> \\s-\\v\\$ ', 'HOSTNAME': 'mock'} and
shell False
octave_cmd: invalid option -- 'n'
error: Unknown option n in octave_cmd()
error: line 31: %octave_pkg_install
%install
%octave_pkg_install
%octave_pkg_install \
mkdir -p %{buildroot}%{octprefix} \
mkdir -p %{buildroot}%{octarchprefix} \
%octave_cmd
pkg("prefix","%{buildroot}%{octprefix}","%{buildroot}%{octarchprefix}");pkg("global_list",fullfile("%{buildroot}%{octshareprefix}","octave_packages"));pkg("local_list",fullfile("%{buildroot}%{octshareprefix}","octave_packages"));pkg("install","-nodeps","-verbose","%{_builddir}/%{buildsubdir}/build/%{octpkg}-%{version}-%{octave_tar_suffix}.tar.gz");unlink(pkg("local_list"));unlink(pkg("global_list"));
\
if [ -e %{buildroot}%{octpkgdir}/packinfo/on_uninstall.m ] \
then \
mv %{buildroot}%{octpkgdir}/packinfo/on_uninstall.m
%{buildroot}%{octpkgdir}/packinfo/on_uninstall.m.orig \
fi \
echo "function on_uninstall (desc)" >
%{buildroot}%{octpkgdir}/packinfo/on_uninstall.m \
echo " error ('Can not uninstall %s installed by the redhat package
manager', desc.name);" >>
%{buildroot}%{octpkgdir}/packinfo/on_uninstall.m \
echo "endfunction" >> %{buildroot}%{octpkgdir}/packinfo/on_uninstall.m \
if [ -e
%{_builddir}/%{buildsubdir}/build/%{octpkg}-%{version}/octave-%{octpkg}.metainfo.xml
] \
then \
echo "Found octave-%{octpkg}.metainfo.xml" \
mkdir -p %{buildroot}/%{_datadir}/appdata \
cp -p
%{_builddir}/%{buildsubdir}/build/%{octpkg}-%{version}/octave-%{octpkg}.metainfo.xml
%{buildroot}/%{_datadir}/appdata/ \
appstream-util validate-relax --nonet
%{buildroot}/%{_datadir}/appdata/octave-%{octpkg}.metainfo.xml \
fi \
%{nil}
%octave_cmd() octave -H -q --no-window-system --no-site-file --eval
'%*';
Somehow I think it's grabbing onto the "-nodeps" as an option.
Right, based on a quick look it seems to be caused by the recently
added single/double quoting support for macro arguments.
IIRC there's no support for escaping the quoting currently, on the
notion that if you need to pass single quotes you use double quotes
and vice versa. But this seems to require both...
+ octave -H -q --no-window-system --no-site-file --eval 'pkg build
"-verbose" "-nodeps" /var/tmp/octave-statistics-1.3.0-5.fc25.x86_64
/home/pmatilai/rpmbuild/BUILD/octave-statistics-1.3.0'
I'll have a closer look tomorrow but please file a bug so it gets
properly tracked. Thanks,
- Panu -
Thanks, filed https://bugzilla.redhat.com/show_bug.cgi?id=1481025
The macro argument quoting in rpm is breaking not just a thing or two
but a whole pile.
I've reverted that now and back to the drawing board at upstream,
quoting support for the arguments is necessary but clearly it needs some
different approach.
- Panu -
_______________________________________________
devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
