It's better to warn by default on MSVC (which we were already doing before
we bumped to 1.4.0) than to fail by default on macOS.
Fixes: 51e3e7d9ae ("build: Bump Meson dependency to 1.4.0")
libinotify-kqueue is a library that implements inotify interface in terms of
kqueue/kevent API available on Mac OS and *BSD systems. The original kqueue
backend seems to be a predecessor version of the code that is currently present
in libinotify-kqueue. Under the hood the library implements a sophisticated
filesystem changes detection algorithm that is derived from the glib backend
code.
Updating the native glib kqueue backend requires substantial work, because code
bases have diverged greatly. Another approach is taken, instead. libinotify-kqueue
can serve as a drop-in replacement for Linux inotify API, thus allowing to
reuse the inotify backend code. The compatibility, however, comes at cost, since
the library has to emulate the inotify descriptor via an unix domain socket.
This means that delivering an event involves copying the data into the kernel
and then pulling it back.
The recent libinotify-kqueue release adds a new mode of operation called "direct".
In this mode the socket pipe is replaced with another kqueue that is used to
deliver events via a kevent(EVFILT_USER) call.
Employing the direct mode requires minor changes to the client code compared
to using plain inotify API, but in return it allows for reusing libinotify's
algorithms without a performance penalty. Luckily, all required changes are
consolidated in one file called inotify-kernel.c
This puts us in the best of possible worlds. On one hand we share a lot of code
with glib inotify backend, which is far more thoroughly tested and widely used.
On the other we support a range of non-Linux systems and consolidate the business
logic in one library. I plan to do the same trick for QFileSystemWatcher which
will give us the same behaviour between Gtk and Qt applications.
The glib test suite passes for both old kqueue backend and new libinotify-kqueue
one. However, the AppStream FileMonitor tests are failing with the old backend,
but pass with the new one, so this is still an observable improvement.
Relevant libinotify-kqueue PR: https://github.com/libinotify-kqueue/libinotify-kqueue/pull/19
The option defaults to 'auto' which keeps the current selection behavior, but
also allows user to override the choice.
Also make the chosen backend is reported to the code via CPP defines.
Meson 1.5.1 is available in the fd.o SDK and in Debian testing, so the
glib Meson policy says we can update. Update the minimum only as far as
1.4.0 because we don't yet have a need for 1.5.0.
This allows us to:
- Use file.full_path() to avoid deprecation warnings on str.format(file).
- Set c_std=gnu99,c99 to avoid deprecation warnings with gnu99 on MSVC.
Update all the CI builds to use the latest 1.4.x patch release, 1.4.2.
The FreeBSD runner cannot be updated via `gitlab-ci.yml`, so will be
broken for now.
Similarly, the macOS build will not work unless `-Dc_std=gnu99` is
specified at configure time, due to
https://github.com/mesonbuild/meson/issues/13639.
Because `epoll_create1()` is what the code in `giounix-private.c`
actually uses.
Spotted by Xuntao Chi.
Signed-off-by: Philip Withnall <pwithnall@gnome.org>
Fixes: #3450
GLib now compiles without emitting any float conversion warnings, so
let’s keep it that way.
Signed-off-by: Philip Withnall <pwithnall@gnome.org>
Helps: #3405
Without the -Wno-typedef-redefinition option, clang complains if a typedef
gets redefined in gnu99 mode (since this is officially a C11 feature). This
also happened with old versions of GCC.
So, don't break the build on such warn, since we want to support
toolchains that supports C11 anyways.
Previously, when cleaning up the temporary directory tree created by
passing G_TEST_OPTION_ISOLATE_DIRS, any symbolic links in that tree
would be followed recursively. If the test case has created a symbolic
link in its temporary directory to a directory outside that tree, this
could lead to unexpected data loss; in particular, if the test case
author has (unwisely) created a symbolic link to /, they could lose all
data on the system.
On systems that have the ftw.h header, replace the current rm_rf()
implementation with one that uses nftw() to perform a depth-first
traversal (FTW_DEPTH) without following symbolic links (FTW_PHYS), and
without crossing mount points (FTW_MOUNT) in case a test has mounted
some other filesystem in the temporary directory.
The callback logs any error to the standard error stream, but returns 0
rather than -1 to allow nftw() to keep walking the tree rather than
terminating immediately. Suppose we are trying to clean up the following
tree:
tmpdir/
a/
f/ (directory not readable for some reason)
g/
p
b/
c
d
Since tmpdir/a/f is not readable, we can expect to fail to delete
tmpdir/a/f, tmpdir/a and tmpdir; but it is preferable to (attempt) to
delete the rest of the tree rather than failing outright. The cost is
that three errors will be logged (for tmpdir/a/f, tmpdir/a and tmpdir).
nftw() is part of POSIX.1-2001, SUSv1, and glibc ≥ 2.1, so should be
available on effectively every platform except Windows. (And Windows
does not enable symbolic links by default so the developer error is less
likely to occur there.)
The macOS ftw(3) manpage says:
> These functions are provided for compatibility with legacy code. New
> code should use the fts(3) functions.
fts(3) does not seem to be part of any standard, but it does seem to be
equally widely supported. The Linux manpages do not indicate that
nftw() is deprecated.
Fixes: https://gitlab.gnome.org/GNOME/glib/-/issues/3290
Now systemtap can be enabled by default in distros that use
-Dauto_features=enabled or for developers who already have systemtap
installed, while it's still disabled for developers who do not have
systemtap installed. See #3354
Now dtrace can be enabled by default in distros that use
-Dauto_features=enabled or for developers who already have dtrace
installed, while it's still disabled for developers who do not have
dtrace installed. See #3354
Since we are now building introspection files for GLib while building
GLib, so we want to make sure that we indeed load the freshly-built
DLLs when running g-ir-scanner, so we add the various needed subdirs
(and if needed, subprojects), to set the GI_EXTRA_BASE_DLL_DIRS
envvar so that g-ir-scanner will look for the newly-built GLib DLLs.
Ideally, upstream g-ir-scanner will need to be updated accordingly to do
something similar to what we are doing here, but this is needed until
the time that we require a g-ir-scanner that contains the update.
This will also fix the g-ir-scanner erroring out when there is no
pre-existing GLib on the system, as the needed DLLs are now found.
Related issue: https://gitlab.gnome.org/GNOME/gobject-introspection/-/issues/499
Related MR in G-I:
https://gitlab.gnome.org/GNOME/gobject-introspection/-/merge_requests/458
It's an array containing the list of sanitizers in use, normally it
contains a value, but in some cases may have more than one (e.g.
'address' and 'undefined').
And so use it to avoid repeated checks
Some architecture such as sparc and some flavors of arm needs -latomic
to avoid the following build failure:
gthread-posix.c:(.text+0xda8): undefined reference to `__atomic_compare_exchange_4'
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
This enables `NULL` pointer dereference checking in the compiler. This
isn’t as good as static analysis, but it should hopefully catch some
simple errors without too high a false positive rate.
If the false positive rate is too high to be useful, we can always
disable it again.
Signed-off-by: Philip Withnall <pwithnall@gnome.org>
Helps: #1767
The python interpreter found by `/usr/bin/env python3` is not
necessarily the same installation as the one that's found by meson's
`pymod.find_installation('python')`. This means that even though meson
is checking that the python installation it found includes the
'packaging' module, the scripts might not have access to that module
when run.
For distribution packaging, it's usually desirable to have python script
interpreters be fully specified paths, rather than use `/usr/bin/env`,
to ensure the scripts run using the expected python installation (i.e.
the one where the python 'packaging' dependency is installed).
The easiest way to fix this is to set the script interpreter to the
`full_path()` of the python interpreter found by meson. The specific
python interpreter that will be used can be selected through the use of
a meson machine file by overriding the "python" program. Many
distributions already have this set up using meson packaging helpers.
If our GPollFunc is set to g_poll() then we can optionally use a poll()
alternative with higher precision. ppoll() provides poll() equivalence
but with timeouts in nanoseconds.
With more precise polling timouts, frame clocks and other timing sensitive
APIs are not restricted to a minimum of 1 millisecond timeout.
This will make it easier and more obvious for developers to run them
locally: I'm sure I'm not the only developer who had assumed that
`.gitlab-ci/` is private to the CI environment and inappropriate (or
perhaps even destructive) to run on a developer/user system.
The lint checks are automatically skipped (with TAP SKIP syntax) if we
are not in a git checkout, or if git or the lint tool is missing. They
can also be disabled explicitly with `meson test --no-suite=lint`,
which downstream distributions will probably want to do.
By default, most lint checks are reported as an "expected failure"
(with TAP TODO syntax) rather than a hard failure, because they do not
indicate a functional problem with GLib and there is a tendency for
lint tools to introduce additional checks or become more strict over
time. Developers can override this by configuring with `-Dwerror=true`
(which also makes compiler warnings into fatal errors), or by running
the test suite like `LINT_WARNINGS_ARE_ERRORS=1 meson test --suite=lint`.
One exception to this is tests/check-missing-install-tag.py, which is
checking a functionally significant feature of our build system, and
seems like it is unlikely to have false positives: if that one fails,
it is reported as a hard failure.
run-style-check-diff.sh and run-check-todos.sh are not currently given
this treatment, because they require search-common-ancestor.sh, which
uses Gitlab-CI-specific information to find out which commits are in-scope
for checking.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This is another way to get the file system type from `statvfs()`, newly
added in glibc 2.39
(https://lwn.net/ml/libc-alpha/38790850.J2Yia2DhmK@pinacolada/).
This hasn’t been tested with glibc 2.39 as I don’t have it, but the
change seems fairly straightforward.
Signed-off-by: Philip Withnall <pwithnall@gnome.org>
The previous commit enabled the `/run/mount/utab` monitoring. The problem
is that the `mount-changed` signal can be emitted twice for one mount. One
for the `/proc/mounts` file change and another one for the `/run/media/utab`
file change. This is still not ideal because e.g. the `GMount` objects for
mounts with the `x-gvfs-hide` option are added and immediately removed.
Let's enable the `mnt_monitor_veil_kernel` option to avoid this.
Related: https://github.com/util-linux/util-linux/pull/2725
