The kernel used on some autobuilders for the Debian PA-RISC port
can do "bwrap --ro-bind / / /bin/true", but not the bwrap invocations
made during the actual testing, which fail with "Creating new namespace
failed: Invalid argument". Make the trial bwrap invocation more like
what Flatpak actually does, so that these tests will hopefully be
skipped on such kernels.
Bug-Debian: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=876743
Signed-off-by: Simon McVittie <smcv@debian.org>
This makes it easier to reason about. In the absence of
FLATPAK_SYSTEM_CACHE_DIR, it would use /var/tmp even during tests that
prefer /tmp.
Signed-off-by: Simon McVittie <smcv@collabora.com>
- Replace `if (g_test_verbose ()) g_print("...\n")` with
`g_test_message ("...")`, which prints the message if the test is
either verbose or in TAP mode. Otherwise, post-mortem debugging
of failing tests on an autobuilder is very difficult.
In some cases, since commit f1dbe9bc "tests: Print spawned program
argv in testlibrary" we already had the g_test_message(), but
still had the redundant g_print() too.
- Factor out running a subprocess into a helper function to reduce
duplication.
- Capture invoked subprocesses' stderr and stdout where possible, so we
can log it as diagnostics. Again, this should make it easier to carry
out post-mortem debugging based on autobuilder logs.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This is declared in the header file, but was never actually implemented.
Oops.
If anybody is hit by this issue, they can work around it by using
g_object_get() to get the FlatpakRemote:type property, which this is the
getter for.
Add it to the tests as well, so it gets exercised.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Use g_main_context_iteration() manually in a loop instead; this makes
the termination conditions more obvious. This does not change the
behaviour of the test.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
This adds variable support for collection IDs: they can either be
enabled on the server, on the server and client, or not at all. If
enabled on the server, apps and runtimes are built with collection IDs
and the repository has one set. If enabled on the client, the remote
config is added to the local repository with a collection ID and GPG
verification enabled. They are controlled with
USE_COLLECTIONS_IN_{SERVER,CLIENT}={yes,no}.
These variables are used in the new wrapper tests,
test-repo-collections.sh and test-repo-collections-server-only.sh.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
This isn’t really used internally, but will be used by gnome-software
for when it configures new flatpak remotes.
This is new public API, but is only declared if compiling with
--enable-p2p.
Includes some basic smoketests.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
This adds a new collection-id property which is only enabled if
FLATPAK_ENABLE_P2P is defined. The internal machinery for handling it is
always enabled, to reduce the amount of #ifdef spam.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Clang otherwise complains about the printf() format string not being a
string literal, which is a bit pants.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
The library test previously used the real ~/.cache, while the
library test and the shell-script tests would use the real ~/.config
to look up the XDG user-dirs.dirs. Other home-directory-related code
might have used the real $HOME.
As a general rule, build-time tests should not affect the real home
directory. Debian autobuilders run as a user whose home directory
does not exist, in order to catch packages whose build process could
affect or be affected by the contents of the home directory. This
caused testlibrary to fail when it tried to create that nonexistent
directory, which I think happened while trying to create ~/.cache.
Signed-off-by: Simon McVittie <smcv@debian.org>
This makes the ostree trivial-httpd --autoexit feature work better,
because it seems to exit whenever the root directory changes (i.e. not
only when its deleted).
This means the root dir can't be the repo (because then we can't
update the repo), or the base testdir (because we create files there
too), so instead we make the repo $testdir/repos/test and
$testdir/repos as the httpd root.
In autobuilder environments that deprive the build of capabilities,
it is entirely possible that we have a system bwrap(1) but cannot
run it, for example because CAP_SYS_ADMIN has been excluded from
the capability bounding set. Tell the tests which bwrap we are
going to run, so we can run it in a simpler way and see whether
it works.
Debian's sbuild autobuilder currently suffers from a different
issue in which pivot_root(2) returns EINVAL, possibly caused by
sbuild being chroot-based and so not having the mount point
structure that is required for pivot_root. This avoids the
problematic build-time tests there too; they work on ci.debian.net,
which uses lxc instead of chroots, and in virtual machines.
Because $(BWRAP) might be non-absolute, we need to search PATH for it.
Signed-off-by: Simon McVittie <smcv@debian.org>
An assertion failure that says res is FALSE is a lot less useful
than an assertion failure that says we got a specific GError.
Signed-off-by: Simon McVittie <smcv@debian.org>
