On 5/24/2020 11:56 AM, Jon Turney wrote:
On 21/05/2020 18:07, Ken Brown via Cygwin-apps wrote:
On 5/21/2020 11:48 AM, Jon Turney wrote:
On 21/05/2020 16:13, Ken Brown via Cygwin-apps wrote:
On 5/21/2020 9:24 AM, Jon Turney wrote:
On 20/05/2020 15:50, Ken Brown via Cygwin-apps wrote:
On 5/19/2020 7:04 PM, Ken Brown via Cygwin-apps wrote:
I would like to adopt gimp and related packages. At the moment I'm
having trouble with babl, which is needed for gegl0.4, which is needed
for gimp. The problem involves gobject-introspection.
If I disable introspection, the build works fine. This would be OK,
since babl has been built without introspection for several years. But
then the gegl0.4 build complains about the missing babl introspection
files, so I would have to disable introspection there too, which hasn't
been done in the past.
So my preference is to figure out what the problem is and get the babl
build working with introspection. I'm attaching my cygport file and patch.
Here's the failing command...
/usr/bin/g-ir-scanner -I/usr/include/gobject-introspection-1.0
-I/usr/include/glib-2.0 -I/usr/lib/glib-2.0/include -D_REENTRANT
--no-libtool --namespace=Babl --nsversion=0.1 --warn-all --output
babl/Babl-0.1.gir --c-include=babl.h
'--identifier-filter-cmd=/usr/bin/python3
/tmp/cygbabl/babl-0.1.74-1.x86_64/src/babl-0.1.74/babl/identfilter.py'
-DBABL_IS_BEING_COMPILED
-I/tmp/cygbabl/babl-0.1.74-1.x86_64/src/babl-0.1.74/babl
-I/tmp/cygbabl/babl-0.1.74-1.x86_64/src/babl-0.1.74/x86_64-pc-cygwin/babl
-I./. -I../. -I./babl/base/. -I../babl/base/.
--filelist=/tmp/cygbabl/babl-0.1.74-1.x86_64/src/babl-0.1.74/x86_64-pc-cygwin/babl/4170c83@@babl-0.1@sha/Babl_0.1_gir_filelist
--cflags-begin -fno-unsafe-math-optimizations
-Wdeclaration-after-statement -Winit-self -Wmissing-declarations
-Wmissing-prototypes -Wold-style-definition -Wpointer-arith -mmmx -msse
-mfpmath=sse -I./. -I../. -I./babl/base/. -I../babl/base/. --cflags-end
--library babl-0.1
-L/tmp/cygbabl/babl-0.1.74-1.x86_64/src/babl-0.1.74/x86_64-pc-cygwin/babl
--extra-library=m --extra-library=dl --extra-library=lcms2
...and the error message:
g-ir-scanner: link: gcc -o
/tmp/cygbabl/babl-0.1.74-1.x86_64/src/babl-0.1.74/x86_64-pc-cygwin/tmp-introspectCwCaUc/Babl-0.1.exe
-ggdb -O2 -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2
-fstack-protector-strong --param=ssp-buffer-size=4
-fdebug-prefix-map=/tmp/cygbabl/babl-0.1.74-1.x86_64/build=/usr/src/debug/babl-0.1.74-1
-fdebug-prefix-map=/tmp/cygbabl/babl-0.1.74-1.x86_64/src/babl-0.1.74=/usr/src/debug/babl-0.1.74-1
/tmp/cygbabl/babl-0.1.74-1.x86_64/src/babl-0.1.74/x86_64-pc-cygwin/tmp-introspectCwCaUc/Babl-0.1.o
-L. -lbabl-0.1 -lm -ldl -llcms2
-L/tmp/cygbabl/babl-0.1.74-1.x86_64/src/babl-0.1.74/x86_64-pc-cygwin/babl
-Wl,-rpath,/tmp/cygbabl/babl-0.1.74-1.x86_64/src/babl-0.1.74/x86_64-pc-cygwin/babl
-lgio-2.0 -lgobject-2.0 -Wl,--export-all-symbols -lgmodule-2.0 -lglib-2.0
-lintl
ERROR: can't resolve libraries to shared libraries: babl-0.1
I don't understand the error message, because the command line contains
-L/tmp/cygbabl/babl-0.1.74-1.x86_64/src/babl-0.1.74/x86_64-pc-cygwin/babl
and that directory contains libbabl-0.1.dll.a and cygbabl-0.1-0.dll. I
even tried adding that directory to my PATH to make sure the right
cygbabl-0.1-0.dll would be found, but that didn't help.
This might possibly be related to the problem described in the comment for:
https://github.com/mesonbuild/meson/pull/2880/commits/8a27c08b05e4537d5061d30ddd8aad9dc52cf1c4
Yeah, this looks extremely plausible, there seem to be no GObject derived types
in babl's interface.
It might be possible to work around this by patching in a dummy GObject derived
type which does nothing.
I'll have to see if I can find my notes about how I debugged this before.
By the way, in case you're wondering why I disabled the building of docs,
it's because I was getting a build failure there too. I don't know if
this is related to the introspection failure. The failing command there is
/usr/bin/meson --internal exe --unpickle
/tmp/cygbabl/babl-0.1.74-2.x86_64/src/babl-0.1.74/x86_64-pc-cygwin/meson-private/meson_exe_env_7bf39b99114d34540b83d26a5d8f097e05882836.dat
cp: target 'docs/index.html.tmp' is not a directory
I don't know why it's not showing me the actual cp command that fails.
I believe it's is an infelicity in meson that it doesn't echo the actual
failing command here.
Noted here:
https://github.com/mesonbuild/meson/pull/3716#issuecomment-395746838
The corresponding information in docs/meson.build is
Reference_html = custom_target('Reference.html',
input : [
'Reference-static.html',
'toc',
index_html_tmp,
],
output: [ 'Reference.html', ],
command: [
env_bin,
'cp', '@INPUT0@', '@OUTPUT@',
'&&', xml_insert, '@OUTPUT@', 'TOC', '@INPUT1@',
'&&', xml_insert, '@OUTPUT@', 'BablBase', '@INPUT2@',
],
build_by_default: true,
)
There are several such custom targets in the file, and for all except this
one, I see the actual cp command in the log. This is the only one for
which meson generates a 'meson --unpickle' command instead of a cp command.
Yeah, I wasn't expecting it to use this method of executing the command
line (storing it in a pickle and then using a python wrapper to execute it)
to be used except on Windows, so I'll have to take a more detailed look at
why that's happening.
So, it's correct that it runs this command indirectly via the pickle, because it
needs to interpose itself to do some PATH manipulation to add the DLL, because
this target has the 'babl-html-dump' tool, which is linked with the DLL, in it's
input (indirectly).
Thanks. FWIW, the recipe for building docs/Reference.html translates to
/usr/bin/env \
cp ../docs/Reference-static.html docs/Reference.html \
&& ../docs/tools/xml_insert.sh docs/Reference.html TOC ../docs/toc \
&& ../docs/tools/xml_insert.sh \
docs/Reference.html BablBase docs/index.html.tmp
This succeeds when run manually in the build directory. So something must
have gone wrong in the pickling/unpickling process.
patching /usr/lib/python3.6/site-packages/mesonbuild/scripts/meson_exe.py
something like this might shed some light:
--- meson_exe.py.bak 2020-05-21 15:01:19.187046500 +0100
+++ meson_exe.py 2020-05-21 15:09:29.485915300 +0100
@@ -57,6 +57,8 @@
['Z:' + p for p in exe.extra_paths] +
child_env.get('WINEPATH', '').split(';')
)
+ print(cmd_args)
+
p = subprocess.Popen(cmd_args, env=child_env, cwd=exe.workdir,
close_fds=False,
stdout=subprocess.PIPE,
OK, now the log shows
cp: target 'docs/index.html.tmp' is not a directory
['/usr/bin/env', 'cp', '../docs/Reference-static.html', 'docs/Reference.html',
'&&',
'/home/kbrown/src/cygpackages/babl/babl-0.1.74-1.x86_64/src/babl-0.1.74/docs/tools/xml_insert.sh',
'docs/Reference.html', 'TOC', '../docs/toc', '&&',
'/home/kbrown/src/cygpackages/babl/babl-0.1.74-1.x86_64/src/babl-0.1.74/docs/tools/xml_insert.sh',
'docs/Reference.html', 'BablBase', 'docs/index.html.tmp']
This does indeed shed some light. If I remove all the commas but leave the
single quotes, the command fails with the same error message as before:
cp: target 'docs/index.html.tmp' is not a directory
If I also remove the single quotes, the command succeeds. I think the problem
is the quotes around the double ampersands, so they are treated as arguments
to the cp command instead of being interpreted by the shell executing the
command.
So, yeah, this is a meson bug, which I will work on (if this command ends up in
the build.ninja, it's executed by ninja with 'sh -c', but if it ends up in a
pickle, it's executed by meson with execve())
Yes, that does seem like a meson bug. But is it also a babl bug to some extent?
When babl puts '&&' in a command argument, it's assuming that the command will
be executed by 'sh -c'.
I have very little experience with meson. Have you ever seen this issue in
other projects that use meson?
(Ofc, everything behaves differently if we are building on Windows, and
CreateProcess() is used in both places, so allowing shell control operators here
is not very portable)