Arch Linux User Repository

Sorry for the delay. I've been rather busy recently.

• I'm not sure if tmp would work because the directory it normally a mounted tmpfs.

• Yes, I agree that using a mounted home isn't a bad setup at all. I just assumed that it's not very common.

• "Expected failure" has a different semantic I think. It means that the test intentionally feeds an invalid input and expects the code to detect the error and report it as some sort of failure. It's not the case for the trash test.

Would the trash test work if it took place in /tmp or /var/tmp? Maybe we could use one of those?

Re. the trashing test failure in a "mounted" location: My /home folder is just a separate btrfs subvolume from /. That doesn't feel like like a very exotic setup to me and building the package in one's home directory also doesn't feel so out of the ordinary.

Another option would be to patch to allow the results of this test to go into the "Expected Fail" category. That way the build could continue normally if it takes place in some location where gio trashing is broken.

From my perspective, 2 or 4 seem sensible. Looking at makepkg man page, it does not seem that there is a dedicated build dir readily available. There is BUILDDIR, which allows to customize where makepkg should build.

Maybe we can print a message on test failure and advise the user to use BUILDDIR to specify an appropriate folder?

@greyltc IMO disabling the entire test is not a good choice because it may hide other bugs in the module reported by the test. Actually I've been willing to help work on this but I just couldn't reproduce it previously. Your link to the bug report seems to provide the right direction.

Here are some possible solutions in my mind:

1. Accurately remove the test code that tests something inherently not supported ("trashing on mounts")

2. Given that it's not very common to build the package in a mounted folder, let the test fail but print some useful information and suggest to build in a non-mounted folder.

3. Modify the test code to test trashing only when building in a non-mounted folder.

4. Modify the test code to find a well-known non-mounted folder to test trashing.

Let me know which one seems to be the most appropriate and I can prepare a patch.

The trash test is failing. See below for the relevant part of my testlog.txt
Maybe this is https://bugs.archlinux.org/task/60137 ?
Should we patch to disable this test?

172/256 glib:gio / trash                        FAIL     0.27 s (killed by signal 6 SIGABRT)

--- command ---
GIO_LAUNCH_DESKTOP='/home/[redacted]/.cache/yay/glib2-static/src/build/gio/gio-launch-desktop' G_TEST_SRCDIR='/home/[redacted]/.cache/yay/glib2-static/src/glib/gio/tests' G_TEST_BUILDDIR='/home/[redacted]/.cache/yay/glib2-static/src/build/gio/tests' GIO_MODULE_DIR='' /home/[redacted]/.cache/yay/glib2-static/src/build/gio/tests/trash --tap
--- stdout ---
# random seed: R02S3787db81b29fedd03c209d37337d06ec
1..2
# Start of trash tests
# Bug Reference: https://gitlab.gnome.org/GNOME/glib/issues/251
# File: /tmp/test-trashW1Q6YZ (parent st_dev: 48)
# Home: /home/[redacted] (st_dev: 49)
# Mount: /tmp
# Error: Trashing on system internal mounts is not supported
ok 1 /trash/not-supported
# Bug Reference: https://gitlab.gnome.org/GNOME/glib/issues/1522
Bail out! GLib-GIO:ERROR:../glib/gio/tests/trash.c:113:test_trash_symlinks: 'target_mount' should not be NULL
--- stderr ---
**
GLib-GIO:ERROR:../glib/gio/tests/trash.c:113:test_trash_symlinks: 'target_mount' should not be NULL
-------

With version 2.58.2-3 makepkg fails for me at check().

The corresponding lines from testlog.txt are the following:

 62/256 glib:glib / search-utils                FAIL     0.17 s (killed by signal 6 SIGABRT)

--- command ---
MALLOC_PERTURB_='107' G_TEST_SRCDIR='/home/user/.cache/yay/glib2-static/src/glib/glib/tests' G_TEST_BUILDDIR='/home/user/.cache/yay/glib2-static/src/build/glib/tests' MALLOC_CHECK_='2' G_DEBUG='gc-friendly' /home/user/.cache/yay/glib2-static/src/build/glib/tests/search-utils --tap
--- stdout ---
# random seed: R02S8e7b7c1d50ba0d2d5e2c1321957a09f9
1..1
# GLib-DEBUG: Started
# GLib-DEBUG: 'he' - 'Hello World' should match: OK
# GLib-DEBUG: 'wo' - 'Hello World' should match: OK
# GLib-DEBUG: 'lo' - 'Hello World' should NOT match: OK
# GLib-DEBUG: 'ld' - 'Hello World' should NOT match: OK
# GLib-DEBUG: 'wo' - 'Hello-World' should match: OK
# GLib-DEBUG: 'wo' - 'HelloWorld' should NOT match: OK
# GLib-DEBUG: 'jor' - 'Jörgen' should match: FAILED
Bail out! GLib:ERROR:../glib/glib/tests/search-utils.c:67:test_search: assertion failed: (ok)
--- stderr ---
**
GLib:ERROR:../glib/glib/tests/search-utils.c:67:test_search: assertion failed: (ok)
-------

By doing:

check() {

meson test -C build

}

I also get the error:

117/256 glib:gio / contenttype   FAIL    0.27 s (killed by signal 6 SIGABRT)

This is a log file:

https://snuglinux.pp.ua/txt/testlog.txt

@greyltc I couldn't reproduce the error. I think doing some debugging in your build environment would help. Replace arch-meson with meson and add --buildtype debug then running gdb on the test executable will be possible.

I would suggest posting patches rather than complete a PKGBUILD so that the changed things can be easily figured out.

Here's a fixed up PKGBUILD for the latest version:

https://gist.github.com/greyltc/c379a37366454ae6f6732fb23a80194b

one check failing for me now with

77/261 glib:glib / utf8-pointer                FAIL     5.88 s (killed by signal 6 SIGABRT)