Completely rewrite the FAM file monitor. Major changes:
- now runs in the worker thread
- dispatches events in a threadsafe way via GFileMonitorSource
- uses unix fd source instead of a GIOChannel
- is now simple enough to fit into one short file
This is the bare minimal effort. This seems not to crash immediately,
but it definitely needs some better testing.
The backend is not in good shape. It could use some serious work.
Use the "interesting" value from g_file_monitor_source_handle_event() to
decide if we're currently being flooded by a stream of boring events.
The main case here is when one or more files is being written to and the
change events are all being rate-limited in the GFileMonitor frontends.
In that case, we become "bored" with the event stream and add a backoff
timeout. In the case that it is exactly one large file being written
(which is the common case) then leaving the event in the queue also lets
the kernel perform merging on it, so when we wake up, we will only see
the one event. Even in the case that the kernel is unable to perform
merging, the context switch overhead will be vastly reduced.
In testing, this cuts down on the number of wake ups during a large file
copy, by a couple orders of magnitude (ie: less than 1% of the number of
wake ups).
Return an "interesting" boolean from the event handler function on
GFileMonitorSource.
An event was "interesting" if it will result in a signal actually being
dispatched to the user. It is "uninteresting" if it only hit an
already-dirty rate limiter.
We will use this information to do some backing off in the backends when
faced with a flood of uninteresting events.
We generally assume that an IN_CREATE event is the start of a series of
events in which another process is doing this:
fd = creat (...) -> IN_CREATE
write (fd, ..) -> IN_MODIFY
write (fd, ..) -> IN_MODIFY
close (fd) -> IN_CLOSE_WRITE
and as such, we use the CHANGES_DONE_HINT event after CREATED in order
to show when this sequence of events has completed (ie: when we receive
IN_CLOSE_WRITE when the user closes the file).
Renaming a file into place is handled by IN_MOVED_FROM so we don't have
to worry about that.
There are many other cases, however, where a new file 'appears' in a
directory in its completed form already, and the kernel reports
IN_CREATE. Examples include mkdir, mknod, and the creation of
hardlinks. In these cases, there is no corresponding IN_CLOSE_WRITE
event and the CHANGES_DONE_HINT will have to be emitted by an arbitrary
timeout.
Try to detect some of these cases and report CHANGES_DONE_HINT
immediately.
This is not perfect. There are some cases that will not be reliably
detected. An example is if the user makes a hardlink and then
immediately deletes the original (before we can stat the new file).
Another example is if the user creates a file with O_TMPFILE. In both
of these cases, CHANGES_DONE_HINT will still eventually be delivered via
the timeout.
Remove all event merging and dispatch logic from GFileMonitor. The only
implementation of GFileMonitor outside of glib is in gvfs and it already
does these things properly.
Get rid of GLocalDirectoryMonitor. We will use a single class,
GLocalFileMonitor, for both directory and file monitoring. This will
prevent every single backend from having to create two objects
separately (eg: ginotifydirectorymonitor.c and ginotifyfilemonitor.c).
Introduce GFileMonitorSource as a thread-safe cross-context dispatch
mechanism. Put it in GLocalFileMonitor. All backends will be expected
to dispatch via the source and not touch the GFileMonitor object at all
from the worker thread.
Remove all construct properties from GLocalFileMonitor and remove the
"context" construct property from GFileMonitor. All backends must now
get the information about what file to monitor from the ->start() call
which is mandatory to implement.
Remove the implementation of rate limiting in GFileMonitor and add an
implementation in GLocalFileMonitor. gvfs never did anything with this
anyway, but if it wanted to, it would have to implement it for itself.
This was done in order to get the rate_limit field into the
GFileMonitorSource so that it could be safely accessed from the worker
thread.
Expose g_local_file_is_remote() internally for NFS detection.
With the "is_remote" functionality exposed, we can now move all
functions for creating local file monitors to a proper location in
glocalfilemonitor.c
Port the inotify backend to adjust to the changes above. None of the
other backends are ported yet. Those will come in future commits.
Remove the hardwired 1 second event queue logic from inotify-kernel and
replace it with something vastly less complicated.
Events are now reported as soon as is possible instead of after a
delay.
We still must delay IN_MOVED_FROM events in order to look for the
matching IN_MOVED_TO events, and since we want to report events in order
this means that events behind those events can also be delayed. We
limit ourselves, however:
- no more than 100 events can be delayed at a time
- no event can be delayed by more than 10ms
https://bugzilla.gnome.org/show_bug.cgi?id=627285
Add a new internal constructor for GLocalFile (which itself is private).
This new constructor allows creating a GLocalFile from a dirname and a
basename, assuming that the dirname is already in canonical form and the
basename is a regular basename.
This will be used for creating GLocalFile instances from the file
monitoring code (for signal emissions).
For all of the effort spent ensuring that this algorithm would be
correctly threadsafe, I messed up the order of operations within a
single thread when porting to the new approach.
Fix that up.
Also: fix some overzealous asserting in the testcases. Since shutdown
is now lazy, we can never surely say !is_running at any particular point
in time.
This can be used to query whether the task has completed, in the sense
that it has had a result set on it, and has already – or will soon –
invoke its callback function.
Notifications for this property are emitted immediately after the task’s
main callback, in the same main context as that callback. This allows
for multiple bits of code to listen for completion of the GTask, which
opens the door for blocking on cancellation of the GTask and improved
handling of ‘pending’ behaviour.
https://bugzilla.gnome.org/show_bug.cgi?id=743636
g_socket_client_add_application_proxy() claimed "When the indicated
proxy protocol is returned by the #GProxyResolver, #GSocketClient will
consider this protocol as supported but will not try to find a #GProxy
instance to handle handshaking." But in fact, it did the checks in the
wrong order, so GProxy proxies ended up overriding
application-specified ones. Fix that.
Also, simplify the code a bit by making use of g_hash_table_add() and
g_hash_table_contains().
https://bugzilla.gnome.org/show_bug.cgi?id=733876
Currently, the Windows code use Winsock2-specific APIs to try to emulate
calls such as inet_pton(), inet_ntop() and if_nametoindex(), which may not
do the job all the time. On Vista and later, Winsock2 does provide a
proper implementation for these functions, so we can use them if they exist
on the system, by querying for them during g_networking_init(). Otherwise,
we continue to use the original code path for these, in the case of XP and
Server 2003.
This enables many of the network-address tests to pass on Windows as a
result, when the native Winsock2 implementations can be used.
https://bugzilla.gnome.org/show_bug.cgi?id=730352
Check the IPv6 addresses on Windows, as we need to reject those that have
brackets/ports around them as valid addresses in this form would have been
accepted during the call to g_inet_address_new_from_string ().
https://bugzilla.gnome.org/show_bug.cgi?id=730352
There was a theoretical deadlock between the worker trying to emit a
signal at the same time as we were waiting for it to shutdown the
notification (while holding the lock).
The deadlock was particularly annoying because we didn't really need to
wait for the shutdown and because it wasn't possible to signals to
arrive while waiting for a start. Attempting to deal with start and
stop in an asymmetric way could have lead to other weird situations,
however.
Drop the lock while waiting for the worker thread to start. This means
that we face the possibility of multiple waiters on the cond at the same
time, so we need to make more of a state machine.
https://bugzilla.gnome.org/show_bug.cgi?id=742599
GUnixMountMonitor was not threadsafe before. It was a global singleton
which emitted signals in the first thread that happened to construct it.
Move it to a per-context singleton model where each GMainContext gets
its own GUnixMountMonitor. Monitor for the changes from the GLib worker
thread and dispatch the results to each context with an active monitor.
https://bugzilla.gnome.org/show_bug.cgi?id=742599
Deprecate g_unix_mount_monitor_set_rate_limit() and turn it into a
no-op.
This function doesn't behave as advertised. It only controls rate
limiting for filesystem-based monitors. It has no impact over reporting
mount changes on Linux, for example, because those are based on polling
for changes in /proc (which doesn't use filesystem monitors). It also
has no impact on Mac OS because a library interface is used there.
This was added in https://bugzilla.gnome.org/show_bug.cgi?id=521946 in
order to be used by HAL, which is effectively dead. udisks no longer
uses this code at all.
https://bugzilla.gnome.org/show_bug.cgi?id=742599
This is a singleton, but we have a function called _new() to get it.
What's worse is that the documentation makes no mention of this, and
actually specifically says that a new monitor will be created each time.
https://bugzilla.gnome.org/show_bug.cgi?id=742599
Add a new internal helper called GContextSpecificGroup.
This is a mechanism for helping to maintain a group of context-specific
monitor objects (eg: GAppInfoMonitor, GUnixMountMonitor).
https://bugzilla.gnome.org/show_bug.cgi?id=742599
If someone explicitly calls g_application_quit() then don't attempt to
drain the mainloop of remaining sources.
This allows applications with 100% CPU utilisation to quit reliably.
https://bugzilla.gnome.org/show_bug.cgi?id=744876
We install win32-software/autorun.exe (as test data for mime scanning)
only on UNIX builds, so don't attempt to chmod it on 'make install'
unless we're on UNIX.
I love Emacs keyboard macros, used them to convert the list of
defines cleverly into a list of tests, then iterated and filled in
the necessary constructor arguments.
After ::shutdown, run the mainloop until all pending activity is
handled, before returning from run().
Among other things, this gives a chance for destroyed windows to be
properly withdrawn from the windowing system.
https://bugzilla.gnome.org/show_bug.cgi?id=744876
This is *significantly* more pleasant to use from C (while handling
errors and memory cleanup).
While we're here, change some ugly, leaky code in
tests/desktop-app-info.c to use it, in addition to a test case
in tests/file.c.
https://bugzilla.gnome.org/show_bug.cgi?id=661554
g_application_bind_busy_property() had the restriction that only one
property can be bound per object, so that NULL could be used to unbind.
Even though this is enough for most uses, it is a weird API.
Lift that restriction and add an explicit unbind function.
https://bugzilla.gnome.org/show_bug.cgi?id=744565
Add an implementation of non-thread-emulated async close of a GIOStream
if either of the underlying stream objects support it.
This prevents us from calling close() functions from another thread on
an object that may not be expecting that. It also allows us to skip the
thread entirely in case our objects support a pure async close.
https://bugzilla.gnome.org/show_bug.cgi?id=741630
Add an internal helper to find out if close_async() is implemented via
threads using the default implementation in the base class.
We will use this to decide if we should do a 'pure async' close of a
GIOStream or not.
https://bugzilla.gnome.org/show_bug.cgi?id=741630
delayed_close_free() calls g_object_unref() on a variable that is
expected to possibly contain NULL (as indicated by the fact that the
NULL case is handled in my_slow_close_output_stream_close_async()).
This is dead code right now (due to a bug in GDBus), which is why it
isn't actually causing a failure. It should still be fixed, however.
https://bugzilla.gnome.org/show_bug.cgi?id=743990
GApplication set the prgname to the application's id when it was running
in service mode. This broke with the addition of new --app-id option,
because g_set_prgname() was called before parsing the options. Calling
it after option parsing doesn't work, because GOptionContext sets
prgname to argv[0] unconditionally.
Instead of changing the semantics of GOptionContext, simply remove this
functionality from GApplication. It is very unusual to have the prgname
set to the app id instead of the binary's name and might confuse people
when looking at logs etc.
When overriding local_command_line() from a subclass,
g_option_context_parse() might never be invokded. Thus, continue setting
the prgname to argv[0] in GApplication.
https://bugzilla.gnome.org/show_bug.cgi?id=743933
Balancing g_application_{un,}mark_busy() is non-trivial in some cases.
Make it a bit more convenient by allowing to bind multiple boolean
properties (from different objects) to the busy state. As long as these
properties are true, the application is marked as busy.
https://bugzilla.gnome.org/show_bug.cgi?id=744565
Add g_list_store_insert_sorted() which takes a GCompareDataFunc to
decide where to insert. This ends up being a very trivial function,
thanks to GSequence.
https://bugzilla.gnome.org/show_bug.cgi?id=743927
Make sure that we only match the _get_type() function name by
restricting the regexp to matching [A-Za-z0-9_]. We were matching on .*
before which means that if we had two _get_type() functions appearing on
a single line then we would get everything in between them included (by
the default rule of '*' being greedy).
This affected G_DECLARE_*_TYPE which puts several uses of _get_type()
into a single line.
GListModel is an interface that represents a dynamic list of GObjects.
Also add GListStore, a simple implementation of GListModel that stores
all objects in memory, using a GSequence.
https://bugzilla.gnome.org/show_bug.cgi?id=729351
Currently the only way to set a state hint on an action is through a
subclass; add a g_simple_action_set_state_hint() method so that this
becomes easier for clients that already use GSimpleAction.
https://bugzilla.gnome.org/show_bug.cgi?id=743521
When implementing blocking operations on top of
nonblocking sockets we should always first try to
perform the operation and then if needed handle
EAGAIN and wait with g_socket_wait_condition.
This is an optimization since we avoid calling
wait condition when it is not needed, but most
importantly this fixes hangs on win32 where some
events (in particular FD_WRITE) are only emitted
after the operation fails with EWOULDBLOCK.
https://bugzilla.gnome.org/show_bug.cgi?id=732439https://bugzilla.gnome.org/show_bug.cgi?id=741707
Add a unit test that checks g_socket_new_from_fd by creating
a gsocket, obtaining its fd, duplicating the fd and then creating
a gsocket from the new fd. This shows a hang on win32 since the
gsocket created from the fd never receives the FD_WRITE event
because we wait for the condition without first trying to write
and windows signals the condition only after a EWOULDBLOCK error.
https://bugzilla.gnome.org/show_bug.cgi?id=741707
We were asking for properties on NM's dbus interface, but if NM is not
running then there won't be any. Check if the name has an owner before
doing anything to it.
https://bugzilla.gnome.org/show_bug.cgi?id=741653
In g_file_make_directory_with_parents(), the my_error variable is used
for several different purposes throughout the whole function, not all of
which are obvious. This explains the situation with some comments.
https://bugzilla.gnome.org/show_bug.cgi?id=719455
Allows sending of multiple messages (packets, datagrams)
in one go using sendmmsg(), thus drastically reducing the
number of syscalls when sending out a lot of data, or when
sending out the same data to multiple recipients.
https://bugzilla.gnome.org/show_bug.cgi?id=719646
Fix two problems:
1) If g_socket_service_stop is called before the accept call is requeued,
then the reference count won't decrease and this code will hang forever:
while (G_OBJECT (service)->ref_count == ref_count)
g_main_context_iteration (NULL, TRUE);
2) Sometimes the testcase fails (maybe 1 in 200 times for me):
GLib-GIO:ERROR:socket-listener.c:73:connection_cb: assertion failed
(G_OBJECT (service)->ref_count == 2): (3 == 2)
Aborted (core dumped)
The problem is that depending on ordering, cancellation of the async
listener can require further main context iterations before it releases
the reference on the socket service. Furthermore, in some cases, it
requires at least one iteration.
https://bugzilla.gnome.org/show_bug.cgi?id=712570
Add a property to GNetworkMonitor indicating the level of network
connectivity: none/local, limited, stuck behind a portal, or full.
The default implementation just returns none or full depending on the
value of is-available.
https://bugzilla.gnome.org/show_bug.cgi?id=664562
Add G_IO_ERROR_CONNECTION_CLOSED as an alias for
G_IO_ERROR_BROKEN_PIPE, and also return it on ECONNRESET.
It doesn't really make sense to try to distinguish EPIPE and
ECONNRESET at the GLib level, since the exact choice of which error
gets returned in what conditions depends on the OS. Given that, we
ought to map the two errors to the same value, and since we're already
mapping EPIPE to G_IO_ERROR_BROKEN_PIPE, we need to map ECONNRESET to
that too. But the existing name doesn't really make sense for sockets,
so we add a new name.
https://bugzilla.gnome.org/show_bug.cgi?id=728928
This is a convenience method for creating a GNetworkAddress which is
guaranteed to return IPv4 and IPv6 loopback addresses. The program
cannot guarantee that 'localhost' will resolve to both types of
address, so programs which wish to connect to a local service over
either IPv4 or IPv6 must currently manually create an IPv4 and another
IPv6 socket, and detect which of the two are working. This new API
allows the existing GSocketConnectable machinery to be used to
automate that.
Based on a patch from Philip Withnall.
https://bugzilla.gnome.org/show_bug.cgi?id=732317
g_settings_has_signal_handlers() checks whether any of the signals has
pending handlers. However, g_signal_has_handler_pending() matches on
exact detail, even when passing 0. Subscribing to one of GSettings'
signals with a detail will fail this check and never connect to the
backend.
Fix this by calling has_handler_pending() with the key as detail as
well.
https://bugzilla.gnome.org/show_bug.cgi?id=740848
