The way I read gtk4 and graphene source code, what the recipe claims
is not true: you can disable introspection in graphene, and it would
still build and install graphene-gobject.pc which is what gtk4 is
looking for. If the needed piece is not installed, then you need to
get to the bottom of why, e.g. specific source code lines where that
decision is made.

That's not really true. If no patches are applied and g-i is disabled, graphene fails in do_compile:

[75/75] /home/kubiznak/projects/poky2/build/tmp/work/core2-64-poky-linux/graphene/1.10.8-r0/recipe-sysroot/usr/bin/g-ir-compiler-wrapper src/Graphene-1.0.gir --output src/Graphene-1.0.typelib --includedir=/usr/share/gir-1.0
FAILED: src/Graphene-1.0.typelib
/home/kubiznak/projects/poky2/build/tmp/work/core2-64-poky-linux/graphene/1.10.8-r0/recipe-sysroot/usr/bin/g-ir-compiler-wrapper src/Graphene-1.0.gir --output src/Graphene-1.0.typelib --includedir=/usr/share/gir-1.0
Could not find GIR file 'GObject-2.0.gir'; check XDG_DATA_DIRS or use --includedir
error parsing file src/Graphene-1.0.gir: Failed to parse included gir GObject-2.0

The way I read gtk4 and graphene source code, what the recipe claims
is not true: you can disable introspection in graphene, and it would
still build and install graphene-gobject.pc which is what gtk4 is
looking for. If the needed piece is not installed, then you need to
get to the bottom of why, e.g. specific source code lines where that
decision is made.

The way I read gtk4 and graphene source code, what the recipe claims
is not true: you can disable introspection in graphene, and it would
still build and install graphene-gobject.pc which is what gtk4 is
looking for. If the needed piece is not installed, then you need to
get to the bottom of why, e.g. specific source code lines where that
decision is made.

This issue was actually caused by missing host dependencies. Shame on me...

I'm slowly progressing with the patch but always end up at the principal issue of hard dependency on g-i being enabled.

For example, graphene recipe does not reflect on "gobject-introspection-data" being/not being in distro features, claiming gtk4 requires build with introspection. Graphene can be patched to be buildable under both circumstances but gtk4 then (unsurprisingly) fails in do_configure step due to missing graphene-gobject-1.0 if g-i is disabled.

I don't think gtk4 can be patched to build with g-i disabled, do you?

If you agree with me, how can we proceed with the fact that some packages (namely gtk4) basically cannot be built without the feature, while still allowing to disable that feature? Is it really necessary to have everything buildable with g-i disabled? Apparently, it's not buildable now.

Is the `bitbake world` command guaranteed to succeed for every commit in the repository? In my case, I end up with failures even with _default_ setup. My point is whether this has to be an issue on my machine (e.g. native tools?), or I should just try a different commit.


Build Configuration:
BB_VERSION = "2.2.0"
BUILD_SYS = "x86_64-linux"
NATIVELSBSTRING = "universal"
TARGET_SYS = "x86_64-poky-linux"
MACHINE = "qemux86-64"
DISTRO = "poky"
DISTRO_VERSION = "4.1"
TUNE_FEATURES = "m64 core2"
TARGET_FPU = ""
meta
meta-poky
meta-yocto-bsp = "HEAD:2c6f0b9228b47459dd1b67d578792cd017006128"
...
ERROR: rust-1.66.0-r0 do_compile: ExecutionError('/home/kubiznak/projects/poky/builds/test-gi/tmp/work/core2-64-poky-linux/rust/1.66.0-r0/temp/run.do_compile.1080378', 1, None, None)
ERROR: Logfile of failure stored in: /home/kubiznak/projects/poky/builds/test-gi/tmp/work/core2-64-poky-linux/rust/1.66.0-r0/temp/log.do_compile.1080378
NOTE: recipe rust-1.66.0-r0: task do_compile: Failed
ERROR: Task (/home/kubiznak/projects/poky/meta/recipes-devtools/rust/rust_1.66.0.bb:do_compile) failed with exit code '1'

The first error in rust log:

error: /home/kubiznak/projects/poky/builds/test-gi/tmp/work/core2-64-poky-linux/rust/1.66.0-r0/rustc-1.66.0-src/build/x86_64-unknown-linux-gnu/stage1-rustc/release/deps/libunic_langid_macros_impl-09018f99e8b09dcb.so: undefined symbol: _Ungind_Resume
--> /home/kubiznak/projects/poky/builds/test-gi/tmp/work/core2-64-poky-linux/rust/1.66.0-r0/rustc-1.66.0-src/vendor/unic-langid-macros/src/lib.rs:5:9
|
5 | pub use unic_langid_macros_impl::langid;
| ^^^^^^^^^^^^^^^^^^^^^^^

Full logs attached.


Petr

________________________________
From: Alexander Kanavin <alex.kanavin@...>
Sent: Wednesday, January 4, 2023 1:36:27 PM
To: Petr Kubizňák - 2N
Cc: openembedded-core@...
Subject: Re: [OE-core][PATCH 1/2] gobject-introspection: check for GI_DATA_ENABLED

On Wed, 4 Jan 2023 at 13:24, Petr Kubizňák <kubiznak@...> wrote:

Yes, that's exactly the reason. I'm particularly concerned about dependency on glib, but other dependencies are unnecessary, too.


This change was sufficient to remove the dependency from the target. I am not aware of potential consequences of removing the -native dependencies, but if you think it should also be done, I'm happy to update the patch. Please let me know.

I'd suggest that you remove all three, then set up a plain poky master
build with gobject-introspection-data disabled, and issue bitbake
world.

Component upstreams generally only test the gi-enabled configuration,
so this may reveal incorrect upstream logic when g-i is disabled.

Alex

If you have a bit of time, I'd suggest that you check the failures one

by one. Some of it, like missing glib-* executables seems like a
trivial missing dependency which was previously pulled in indirectly.

Others, like python3-pygobject probably have a hard dependency on g-i.

When the patch is _applied_ and g-i is _disabled_, graphene does build, but some other packages don't:

This shows that these packages depend on gobject-introspection and/or its dependencies no matter whether introspection is enabled or disabled. I don't know these packages so I don't think I'm able to patch them properly. On the other hand, I still believe it is incorrect to pull in the unnecessary dependencies when g-i is disabled. What do you think?

Yes, that's exactly the reason. I'm particularly concerned about dependency on glib, but other dependencies are unnecessary, too.


This change was sufficient to remove the dependency from the target. I am not aware of potential consequences of removing the -native dependencies, but if you think it should also be done, I'm happy to update the patch. Please let me know.

When GI_DATA_ENABLED is 'False' (e.g. because
'gobject-introspection-data' is not in DISTRO_FEATURES),
gobject-introspection should not be added to DEPENDS.