These variables were already (correctly) accessed atomically. The
`volatile` qualifier doesn’t help with that.
Signed-off-by: Philip Withnall <pwithnall@endlessos.org>
Helps: #600
The previous code consumed a larger additional amount of stack space.
That is because it would allocate the temporary buffer for GValues on
the stack with "g_newa (GValue, 1)" and thus the required stack
space grew with the number of arguments. Granted, this is already
a variadic C function, so the caller already placed that many elements
on the stack. For example, on the stack there are the property names
and the pointers to the arguments, which should amount to roughly
O(n_args * 16) (on 64 bit, with pointers being 8 bytes large).
That is not bad, because it means in the previous version the stack space
would grow linear with the already used stack space. However, a GValue is
an additional 24 bytes (on 64 bit), which probably more than doubles the
required stack space. Let's avoid that, by allocating the temporary list
on the heap after a certain threshold. This probably more than doubles the
number of possible arguments before the stack overflows.
Also, previously the heap allocated "params" array only grew one element
per iteration. Of course, it is likely that libc anyway reallocates
the buffers by growing the space exponentially. So realloc(ptr, 1)
probably does not O() scale worse than doubling the buffer sizes ourselves.
However, it seems clearer to keep track of the allocated sizes ourself, and
only call realloc() when we determine that we are out of space.
Especially because we need to update the value pointers on reallocation.
Note that we now require a heap allocation both for the "params" and the
"values" list. Theoretically that could be combined by using one buffer
for both. But that would make the code more complicated.
Now we pre-allocate buffers for 16 elements on the stack. That
is (16 * (16 + 24) bytes (or 640 bytes) on the stack. I think that
is still acceptable.
Two out of three callers pass the count argument from a variable
of type guint. And the third is currently an always positive gint.
We should use the correct integer type that matches the type as it
used otherwise.
Rather than using a mixture of ‘instantiable’ and ‘instantiatable’
everywhere, standardise on the term which is already in the public API.
Signed-off-by: Philip Withnall <pwithnall@endlessos.org>
This was mostly machine generated with the following command:
```
codespell \
--builtin clear,rare,usage \
--skip './po/*' --skip './.git/*' --skip './NEWS*' \
--write-changes .
```
using the latest git version of `codespell` as per [these
instructions](https://github.com/codespell-project/codespell#user-content-updating).
Then I manually checked each change using `git add -p`, made a few
manual fixups and dropped a load of incorrect changes.
There are still some outdated or loaded terms used in GLib, mostly to do
with git branch terminology. They will need to be changed later as part
of a wider migration of git terminology.
If I’ve missed anything, please file an issue!
Signed-off-by: Philip Withnall <withnall@endlessm.com>
The various `g_strdup_printf()` returns values in the implementations of GValue
lcopy_func are runtime checks which could be disabled if one wants and therefore
should be handled as such with g_return_val_if_fail()
Rename the variables involved so that people get a slightly more
obvious critical warning when they try to ref an object which has
already been finalised.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
As with `g_variant_new()` (or any varargs function which takes integer
literals of differing widths), callers need to be careful to ensure
their integer literals have the right width.
Tweak the documentation for `g_object_new()`, `g_object_set()` and
`g_object_get()` to clarify this. The documentation for `g_object_get()`
shows that it is not subject to the same caveats, since it operates on
pointers.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #833
This uses a 32bit hole in the GObject structure on 64bit arches
as a flag field which can be optionally used for some preformance hints.
Currently there is a flag that gets set any time you connect to a signal
on a GObject which is used as early bailout for signal emissions, and using
the flags field instead of a user-data for checking if a GObject is
under construction.
"Uninitialized value" is partially correct, since it has not been
initialized with a type, but it's more precise to say
"zero-initialized value". It is still a programming error to pass a
pointer to uninitialized memory with arbitrary contents as the value.
Signed-off-by: Simon McVittie <smcv@collabora.com>
Using the generic marshaller has drawbacks beyond performance. One such
drawback is that it breaks the stack unwinding from the Linux kernel due
to having unsufficient data to walk past ffi_call_unixt64. That means that
performance profiling by application developers looks grouped among
seemingly unrelated code paths.
Related to GNOME/Initiatives#10
We already have the GType with which the GValue should be initialized,
so requiring an initialized GValue is not really necessary, and it
actually complicates code that wraps GObject, by requiring the retrieval
of the GParamSpec in order to get the property type. Additionally, it
introduces a mostly unnecessary g_value_reset().
We already changed g_object_getv() to allow passing uninitialized
GValues, but this fell through the cracks.
Closes: #737
These have all been documented as deprecated for a long time, but we’ve
never had a way to programmatically mark them as deprecated. Do that
now.
This is based on the list of deprecations from the reverted commit
80fcb1bc2.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Fixes: #638
g_object_set_data() should only ever be used with a small, bounded set
of keys, or the memory usage of the quark lookup table will grow
unbounded. Document that.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Fixes: #682
I'm trying to use `-fsanitize=thread` for OSTree, and some of
these issues seem to go into GLib. Also, the sanitizers work better if
the userspace libraries are built with them too.
This fix is similar to
b6814bb37c
Mixing atomic and non-atomic reads trips TSAN, so let's change the
assertions to operate on the local values returned from atomic
read/writes.
Without this change I couldn't even *build* GLib with TSAN, since we
use gresources during compilation, which uses GSubprocess, which hits
this code.
(Minor review fixes made by Philip Withnall <withnall@endlessm.com>.)
https://gitlab.gnome.org/GNOME/glib/issues/1224
An assertion is harder to skip over, and using a g_critical() can give
us a more informative error message.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://gitlab.gnome.org/GNOME/glib/issues/179
The implementation is silently discarding this anyway, and
g_object_unref() is using atomic operations. So this should be safe.
Having this here triggers -Wdiscarded-qualifiers when g_clear_pointer()
is fixed to use __typeof__().
The implementation is silently discarding this anyway, and
g_object_unref() is using atomic operations. So this should be safe.
Having this here triggers -Wdiscarded-qualifiers when g_clear_pointer()
is fixed to use __typeof__().
There is no transfer annotation that can express transfer semantics of
g_object_new_with_properties in general. When GInitiallyUnowned object
is constructed the introspection data will be incorrect.
Mark it with skip annotation.
https://bugzilla.gnome.org/show_bug.cgi?id=795025
Currently, g_object_ref() and g_object_ref_sink() return a
gpointer which can mask issues when assigning to fields or
returning from a function.
To help catch these type of programming errors, we can propagate
the type of the parameter through the function call on GCC
using the typeof() C language extension.
This will cause offending code to have a warning, but will
continue to be source and binary compatible.
This is only enabled when GLIB_VERSION_MAX_ALLOWED is 2.56 or greater.
https://bugzilla.gnome.org/show_bug.cgi?id=790697
Where we were already treating GHashTables as sets, modify them to use
the set-specific APIs g_hash_table_add() and g_hash_table_contains(), to
make that usage more obvious and less prone to being broken.
Heavily based on patches by Garrett Regier <garrettregier@gmail.com>.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
https://bugzilla.gnome.org/show_bug.cgi?id=749371
This was duplicated also in g_object_interface_install_property().
Now, validations specific to classes happen in
validate_and_install_class_property() - specifically, the checks for
the presence of the get_property() and set_property() methods.
https://bugzilla.gnome.org/show_bug.cgi?id=787551
It's unnecessary, and only adds visual noise; we have been fairly
inconsistent in the past, but the semi-colon-less version clearly
dominates in the code base.
https://bugzilla.gnome.org/show_bug.cgi?id=669355
This is going to be checked again by g_object_new_with_properties()
and g_object_new_valist() anyway, so might just as well leave it
to those functions to do the check and only do it once. It doesn't
matter which function emits the critical warning in the end either,
as one has to look at a stack trace to find out what code triggered
it in any case.
https://bugzilla.gnome.org/show_bug.cgi?id=780908
g_object_newv uses a GParameter as argument. Since GParameter
is deprecated due to this type is not introspectible,
g_object_newv is deprecated now.
https://bugzilla.gnome.org/show_bug.cgi?id=709865
g_object_new_with_properties is an alternative to g_object_newv.
The last one, takes an array of GParameter. However, GParameter
is a rarely used type and this type is not introspectible, so
it will not work properly in bindings.
https://bugzilla.gnome.org/show_bug.cgi?id=709865
If we have an input parameter (or return value) we need to use (nullable).
However, if it is an (inout) or (out) parameter, (optional) is sufficient.
It looks like (nullable) could be used for everything according to the
Annotation documentation, but (optional) is more specific.
The C spec leaves conditional evaluation inside a macro expansion as
undefined behaviour. This means we cannot use constructs like:
GOBJECT_IF_DEBUG(OBJECTS, {
...
#ifdef BLAH
...
#endif
...});
Because compilers are entirely justified to ignore the conditional, or,
like in the case of MSVC, error out.
https://bugzilla.gnome.org/show_bug.cgi?id=769504
historically, DEBUG_CODE(gtype.c) and IF_DEBUG(gobject.c, gsignal.c)
macros are used to support debugging messages about object bookkeeping
and signal emission.
DEBUG_CODE has never been used in gtype.c. IF_DEBUG, when used, must be
accompanied by an extra #ifdef G_ENABLE_DEBUG. this is cumbersome.
this patch add a new macro GOBJECT_IF_DEBUG based on DEBUG_CODE as
a replacement for both DEBUG_CODE and IF_DEBUG.
https://bugzilla.gnome.org/show_bug.cgi?id=729914
Add various (nullable) and (optional) annotations which were missing
from a variety of functions. Also port a couple of existing (allow-none)
annotations in the same files to use (nullable) and (optional) as
appropriate instead.
Secondly, add various (not nullable) annotations as needed by the new
default in gobject-introspection of marking gpointers as (nullable). See
https://bugzilla.gnome.org/show_bug.cgi?id=729660.
This includes adding some stub documentation comments for the
assertion macro error functions, which weren’t previously documented.
The new comments are purely to allow for annotations, and hence are
marked as (skip) to prevent the symbols appearing in the GIR file.
https://bugzilla.gnome.org/show_bug.cgi?id=719966
Keeping these enabled causes too many people to file
bugs against gobject, and not enough people to send
patches to port away from deprecated properties.
Practically no caller of these functions require atomic behaviour,
but the atomics are much slower than normal operations, which makes
it desirable to get rid of them. We have not done this before because
that would be a break of the ABI.
However, I recently looked into this and it seems that even if the
atomics *are* used for g_clear_* it is not ever safe to use this. The
atomics protects two threads that are racing to free a global/shared
object from freeing the object twice. However, any *user* of the global
object have no protection from the object being freed while in use,
because there is no paired operation the reads and refs the object
as an atomic unit (nor can such an operation be implemented using
purely atomic ops).
So, since nothing could safely have used the atomic aspects of these
functions I consider it acceptable to just remove it.
https://bugzilla.gnome.org/show_bug.cgi?id=733969
Don't emit property deprecation warnings for construct properties that
are being set to their default value during construction, but _do_ emit
them in all cases when the property was explicitly given to
g_object_new().
https://bugzilla.gnome.org/show_bug.cgi?id=732184
By default G_PARAM_DEPRECATED means absolutely nothing. We only emit a
warning if G_ENABLE_DIAGNOSTIC is set to '1' and then, only on sets.
Turn the logic on its head: emit the warning by default, unless
G_ENABLE_DIAGNOSTIC is set to 0. In order to avoid a torrent of output, only
emit a warning once per property name.
https://bugzilla.gnome.org/show_bug.cgi?id=732184
Leave ourselves a little wiggle room: if people install properties after
initialisation then we reserve the right to handle that in a way that
may not be threadsafe.
https://bugzilla.gnome.org/show_bug.cgi?id=698614
Add a flag to prevent the automatic emission of the "notify" signal
during g_object_set_property().
If this flag is set then the class must explicitly emit the notify
for themselves. This is already standard practice on most classes, but
we cannot simply remove the existing behaviour because there are surely
many cases where it is needed.
https://bugzilla.gnome.org/show_bug.cgi?id=731200
Construct properties are always set during construction.
It makes no sense to warn about this even if the property
is marked as deprecated; the deprecation warning should
only be issues for explicit uses of the property after
construction.
https://bugzilla.gnome.org/show_bug.cgi?id=730045
Although returning NULL from constructor is strongly discouraged, some
old libraries need to keep doing it for ABI-compatibility reasons.
Given this, it's rude to forbid finalization from within
constructor(), since it would otherwise work correctly now anyway (and
the critical when returning NULL should discourage any new uses of
returning NULL from constructor()).
https://bugzilla.gnome.org/show_bug.cgi?id=661576
Since we are no longer using sgml mode, using /* */ to
escape block comments inside examples does not work anymore.
Switch to using line comments with //
This is really just a very crude and limited conditional breakpoint.
Update the documentation to explain conditional breakpoints in
gdb instead. Also, remove the link to refdbg, which appears dead.
https://bugzilla.gnome.org/show_bug.cgi?id=719687
The signals queued while notify is frozen are emitted in
reverse order, while omitting duplicates. The lack of documentation
for this was pointed out in
https://bugzilla.gnome.org/show_bug.cgi?id=607016
Rather than keeping a global list of objects that are being
constructed, use qdata on the object itself like we do with several
other properties now.
https://bugzilla.gnome.org/show_bug.cgi?id=661576
If a constructor() implementation created an object but then unreffed
it rather than returning it, that object would get left on the
construction_objects list, which would cause problems later when that
memory location got reused by another object.
"Fix" this by making it fail intentionally, and add a test for it (and
for the normal, working singleton case).
https://bugzilla.gnome.org/show_bug.cgi?id=661576
Just like g_object_notify, check for a zero ref_count in
g_object_notify_by_pspec and leave if it is 0.
This allows using functions in ->finalize() that possibly also
notify a property change on the object. Previously,
this resulted in an error from g_object_ref.
https://bugzilla.gnome.org/show_bug.cgi?id=705570
We have turned up enough cases of this being done (including GTK API
allowing apps to do this to GtkSettings well after it has been
instantiated) that it is clear that we cannot really break this feature
while claiming to be backwards compatible.
For that reason, it becomes a warning rather than a critical (ie: it is
still well-defined behaviour, but you are discouraged from doing it).
The intention is to keep this feature for at least the next while.
A given GObjectClass will be able to avoid using GParamSpec pool for as
long as you don't install properties after init. If you do that, you
will get a warning and we will devolve to using GParamSpecPool.
https://bugzilla.gnome.org/show_bug.cgi?id=698614
GObject has previously allowed installing properties after class_init
has finished running. This means that you could install some of your
own properties on G_TYPE_OBJECT, for example, although they wouldn't
have worked properly.
A previous patch asserted that this was not true and we had to revert it
because it broke the shell. Instead of reverting, we should have used a
critical, so do that now.
Complaints go to this bug:
https://bugzilla.gnome.org/show_bug.cgi?id=698614
Back in the far-off twentieth century, it was normal on unix
workstations for U+0060 GRAVE ACCENT to be drawn as "‛" and for U+0027
APOSTROPHE to be drawn as "’". This led to the convention of using
them as poor-man's ‛smart quotes’ in ASCII-only text.
However, "'" is now universally drawn as a vertical line, and "`" at a
45-degree angle, making them an `odd couple' when used together.
Unfortunately, there are lots of very old strings in glib, and also
lots of new strings in which people have kept up the old tradition,
perhaps entirely unaware that it used to not look stupid.
Fix this by just using 'dumb quotes' everywhere.
https://bugzilla.gnome.org/show_bug.cgi?id=700746
There is some code in the wild (like in gnome-session) that does this
from its custom _constructor() implementation:
{
GObject *obj;
obj = ((chain up));
if (!object_is_viable (obj))
{
g_object_unref (obj);
return NULL;
}
else
return obj;
}
This has never been a valid use of GObject and this code has always
caused memory to be leaked[1] by growing the construction_objects list.
The ability to legitimately return NULL from a constructor was exactly
the reason that we created GInitable, in fact.
That doesn't change the fact that the g_object_new() rewrite will crash
in this case, so instead of doing that, let's emit a critical and avoid
the crash. This will allow people to upgrade their GLib without also
upgrading their gnome-session. Meanwhile, people can fix their broken
code.
[1] not in the strictest sense of the word, because it's still reachable
Make a number of improvements to g_object_new():
- instead of looking up the GParamSpec for the named property once in
g_object_new() (in order to collect) and then again in g_object_newv
(when actually setting the property), g_object_new_internal() is a
new function that takes the GParamSpec on the interface to avoid the
second lookup
- in the case that ->constructor() is not set, we need not waste time
creating an array of GObjectConstructParam to pass in. Just directly
iterate the list of parameters, calling set_property() on each.
- instead of playing with linked lists to keep track of the construct
properties, realise that the number of construct properties that we
will set is exactly equal to the length of the construct_properties
list on GObjectClass and the only thing that may change is where the
value comes from (in the case that it was passed in)
This assumption was already implicit in the existing code and can be
seen from the sizing of the array used to hold the construct
properties, but it wasn't taken advantage of to make things simpler.
- instead of allocating and filling a separate array of the
non-construct properties just re-iterate the passed-in list and set
all properties that were not marked G_PARAM_CONSTRUCT (since the ones
that were construct params were already used during construction)
- use the new g_param_spec_get_default_value() API instead of
allocating and setting the GValue for each construct property that
wasn't passed from the user
Because we are now iterating the linked list of properties in-order we
need to append to that list during class initialising instead of
prepending.
These changes show a very small improvement on the simple-construction
performance testcase (probably just noise) and they improve the
complex-construction case by ~30%.
Thanks to Alex Larsson for reviews and fixes.
https://bugzilla.gnome.org/show_bug.cgi?id=698056
This reverts commit ddb0ce1421.
Colin's smoke testing has found issues in at least gjs and
gnome-settings-daemon. We'll need to see if we can address those.
GObject has previously allowed installing properties after class_init
has finished running. This means that you could install some of your
own properties on G_TYPE_OBJECT, for example, although they wouldn't
have worked properly.
Prevent this from happening. Require that all properties are installed by
the time class_init has finished.
Complaints go to this bug:
https://bugzilla.gnome.org/show_bug.cgi?id=698614
This reverts commit 028d4a03f2.
I thought that we would be able to get away with this incompatible
change but it appears to impact far too much existing code. The only
thing we can do is revert.
https://bugzilla.gnome.org/show_bug.cgi?id=688596
Move the constructed() call to happen after all of the properties are
set (not just the construct properties).
This is an incompatible change but we are making it under the belief
that it should be safe. If this change impacts you in a negative way
please comment on the bug.
https://bugzilla.gnome.org/show_bug.cgi?id=685733
Expand the documentation for g_object_[freeze|thaw]_notify() to explain that
it deduplicates “notify” signals emitted by frozen objects, so that at most
one signal is emitted per property.
https://bugzilla.gnome.org/show_bug.cgi?id=676937
Transparent access to a weak pointer from the thread performing the
weak -> strong conversion is incompatible with thread-safety: that
thread will have to do something special. This is GNOME#548954.
https://bugzilla.gnome.org/show_bug.cgi?id=548954
We were previously preventing implementations of an interface from
specifying G_PARAM_CONSTRUCT for a property of that interface if the
interface didn't specify it itself (or was readonly).
This is something that should only interest the implementation, so we
remove this restriction.
This allows 6 new possible override scenarios:
- writable -> writable/construct
- writable -> readwrite/construct
- readwrite -> readwrite/construct
- writable/construct-only -> writable/construct
- writable/construct-only -> readwrite/construct
- readwrite/construct-only -> readwrite/construct
and we update the testcase to reflect this.
https://bugzilla.gnome.org/show_bug.cgi?id=666616
Change the order of the checks so that we hear about the 'biggest'
problem first. Also, stop reporting problems after we report the first
one for a particular property.
Add some comments.
https://bugzilla.gnome.org/show_bug.cgi?id=666616
The property override typecheck was meant to enforce the type on the
overriding property being exactly equal to the type on the interface
property. Instead, g_type_is_a() was incorrectly used.
We could try to enforce equality, but if a property is read-only then it
should be possible for the implementation to type the property with any
subtype of the type specified on the interface (because returning a more
specific type will still satisfy the interface). Likewise, if the
property is write-only then it should be possible for the implementation
to type the property with any supertype.
We implement the check this way.
https://bugzilla.gnome.org/show_bug.cgi?id=666616
Simplify some of the logic in this function.
1) Simplify flag checks as per Colin's suggestions in
https://bugzilla.gnome.org/show_bug.cgi?id=605667
2) Don't repeatedly recheck if class_pspec is NULL.
GObject enforces the following restrictions on property overrides:
- must only add abilities: if the parent class supports
readability/writability then the subclass must also support them.
Subclasses are free to add readability/writability.
- must not add additional restrictions: if the parent class doesn't
have construct/construct-only restrictions then the subclass must
not add them. Subclasses are free to remove restrictions.
The problem with the previous implementation is that the check against
adding construct/construct-only restrictions was being done even if the
property was not previously writable. As an example:
"readable" and "writable only on construct"
was considered as being more restrictive than
"read only".
This patch tweaks the check to allow the addition of
construct/construct-only restrictions for properties that were
previously read-only and are now being made writable.
https://bugzilla.gnome.org/show_bug.cgi?id=666615
Either g_type_register_static_simple (used by G_DEFINE_TYPE_EXTENDED)
and G_IMPLEMENT_INTERFACE use automatic variables for GTypeInfo and
GInterfaceInfo structs, while tutorials and source code often use
static variables. This commit consistently adopts the former method.
https://bugzilla.gnome.org/show_bug.cgi?id=600161
When the 'conditional' parameter is TRUE, the queue will only be frozen
(ie: have its freeze count increase by one) if it is already frozen.
This will allow us to avoid a freeze-notify-thaw in the case that we
just want to notify on a single property.
Another approach may have been to add an is_frozen() type call and avoid
even increasing the freeze count at all in this case. Unfortunately,
I'm not totally sure what is the exact expected semantics of
simultaneous notifications in multiple threads and this may interact
badly with someone freezing or thawing in between our check and
emission.
Lift the check-if-READABLE and redirect-target logic from out of
g_object_notify_queue_add() into its own function, get_notify_pspec().
Use that function at the site of our two calls to
g_object_notify_queue_add().
This was done as a separate file before, and #include brought it into
gobject.c. That's a bit mad, so stop doing that.
Unfortunately, the insanity steps up a level: gobjectnotifyqueue.c is
installed in the public include dir, so we can't just get rid of it
entirely.
Similar to G_PARAM_DEPRECATED. It will warn only for users of the
signals, so a signal can still be emited without warning, for
compatibility reasons.
Apparently, there is no way user flags could have been used before,
so that shouldn't break anyone.
https://bugzilla.gnome.org/show_bug.cgi?id=663581
Some links were broken due to typos, because functionality was removed
in GLib 2.0 or for various other reasons. Fix up as many of them as is
reasonable.
