GNetworkAddress was allowing IPv6 scope ids in g_network_address_new()
/ g_network_address_parse(), but not in g_network_address_parse_uri().
Fix that.
Part of https://bugzilla.gnome.org/show_bug.cgi?id=669724
Convert {glib,gobject,gio}/tests to use the automake TAP driver
and test harness instead of gtester. To do so, we add a glib-tap.mk
that provides the same interface as glib.mk, except for the
reporting and coverage testing functionality. Eventually, we may
want to replace glib.mk with it. I've not yet converted the
toplevel tests/ directory, since it mixes gtestutils tests with
other binaries.
https://bugzilla.gnome.org/show_bug.cgi?id=692125
With UDP sockets, g_socket_bind() with allow_reuse=TRUE on Linux
behaved in a way that the documentation didn't suggest, and that
didn't match other OSes. (Specifically, it allowed binding multiple
multicast sockets to the same address.)
Since this behavior is useful, and since allow_reuse didn't have any
other meaning with UDP sockets, update the docs to reflect the Linux
behavior, and make it do the same thing on non-Linux.
https://bugzilla.gnome.org/show_bug.cgi?id=689245
When running a task in a thread, GTask may still be internally holding
a ref on the task in that thread even after the callback is called in
the original thread (depending on thread scheduling). Fix the test to
handle that by using a weak notify that signals a GCond, and wait for
that GCond from the main thread. (And add a corresponding check to
test_return_on_cancel().)
https://bugzilla.gnome.org/show_bug.cgi?id=705152
Add a missing Name entry, and add a terminal ; to the Actions
entry in org.gtk.test.dbusappinfo.desktop. desktop-file-validate
still contains about the DBusActivatable entry and about the
missing Exec entries. The former will go away when desktop-file-validate
gets updated for the latest spec revision.
Add a fairly realistic testcase that ensures that GDesktopAppInfo with
DBusActivatable=true can successfully talk to GApplication for a variety
of purposes.
https://bugzilla.gnome.org/show_bug.cgi?id=699259
For some time, the desktop file specification has supported "additional
application actions". This is intended to allow for additional methods
of starting an app, such as a mail client having a "Compose New Message"
action that brings up the compose window instead of the folder list.
This patch adds support for this with a relatively minimal API.
In the case that the application is a GApplication and DBusActivatable,
desktop actions are translated into GActions that have been added to the
application with g_action_map_add_action(). This more or less closes
the loop on being able to activate an application with an action
invocation (instead of 'activate').
https://bugzilla.gnome.org/show_bug.cgi?id=664444
Add a new type of GAction that represents the value of a property on an
object. As an example, this might be used on the "visible-child-name"
property of a GtkStack.
https://bugzilla.gnome.org/show_bug.cgi?id=703270
It's possible to get a org.freedesktop.Properties.GetAll call even if we
have no readable properties in the introspection, in which case we
should return the empty list in the usual way.
We should certainly _not_ be dispatching to the method call handler of
an interface which has no properties (since it will not be expecting
this).
Add a check to make sure that there is at least one readable property
before assuming that a NULL get_property handler implies that we want to
handle properties asynchronously.
Add a testcase that was failing before the change and works after it.
https://bugzilla.gnome.org/show_bug.cgi?id=703437
The existing advice in the documentation to "simply" register the
"org.freedesktop.DBus.Properties" interface if you want to handle
properties asynchronously is pretty unreasonable. If you want to handle
this interface you have to deal with all properties for all interfaces
on the path, and you have to do all of the checking for yourself. You
also have to provide your own introspection data.
Introduce a new convention for dealing with properties asynchronously.
If the user provides NULL for their get_property() or set_property()
functions in the vtable and has properties registered then the
properties are sent to the method_call() handler. We get lucky here
that this function takes an "interface_name" parameter that we can set
to "org.freedesktop.DBus.Properties".
We also do the user the favour of setting the GDBusPropertyInfo on the
GDBusMethodInvocation for their convenience (for much the same reasons
as they might want the already-available GDBusMethodInfo).
Add a testcase as well as a bunch of documentation about this new
feature.
https://bugzilla.gnome.org/show_bug.cgi?id=698375
OS X's getaddrinfo() only supports IPv6 scope IDs that are interface
names, not numbers. So use if_indextoname() to get the name of an
interface and construct an address using that.
https://bugzilla.gnome.org/show_bug.cgi?id=700123
In the case that HAVE_DBUS_DAEMON was undefined (as in ostree where glib
is built before D-Bus) this test was failing. Move it inside the
HAVE_DBUS_DAEMON block.
Remove the complications that were introduced in an attempt to make the
gsettings and gschema-compile tests function as installed tests. These
tests are designed (in large part for gsettings and entirely for
gschema-compile) to test the in-tree tools and should not be testing the
system versions.
In the future we may want to move the use of the in-tree tools from the
gsettings testcase into the Makefile and install the resulting files,
allowing this testcase to run against those files, installed.
Perform a substantial cleanup of the build system with respect to
building and installing testcases.
First, Makefile.decl has been renamed glib.mk and substantially
expanded. We intend to add more stuff here in the future, like canned
rules for mkenums, marshallers, resources, etc.
By default, tests are no longer compiled as part of 'make'. They will
be built when 'make check' is run. The old behaviour can be obtained
with --enable-always-build-tests.
--disable-modular-tests is gone (because tests are no longer built by
default). There is no longer any way to cause 'make check' to be a
no-op, but that's not very useful anyway.
A new glibtests.m4 file is introduced. Along with glib.mk, this
provides for consistent handling of --enable-installed-tests and
--enable-always-build-tests (mentioned above).
Port our various test-installing Makefiles to the new framework.
This patch substantially improves the situation in the toplevel tests/
directory. Things are now somewhat under control there. There were
some tests being built that weren't even being run and we run those now.
The long-running GObject performance tests in this directory have been
removed from 'make check' because they take too long.
As an experiment, 'make check' now runs the testcases on win32 builds,
by default. We can't run them under gtester (since it uses a pipe to
communicate with the subprocess) so just toss them in TESTS. Most of
them are passing on win32.
Things are not quite done here, but this patch is already a substantial
improvement. More to come.
This should be the last users that need to be ported.
For some of the oldschool non-gtester-ified tests, we call g_test_init()
from main() because it is necessary in order to use
g_test_build_filename().
Since this feature is so utterly automake-centric, we may as well be
using the same terminology as automake itself (ie: although it's
BUILT_SOURCES, it's DIST_EXTRA, not DISTED).
Also add some comments to the enum explaining that these terms are
really corresponding directly to the automake terms.
https://bugzilla.gnome.org/show_bug.cgi?id=549783
It's a recipe for race conditions and error; on some hardware
architectures one thread isn't guaranteed to see the results
of writes from another thread without a cache flush.
https://bugzilla.gnome.org/show_bug.cgi?id=700855
The test /gdbus/connection/large_message waits for a dbus name to appear.
The dbus name is created by a another process executed in the background.
If for some reason this fails, the test will likely wait forever.
This will avoid this situation by making the test fail if the dbus service
has not appeared after 10 seconds.
https://bugzilla.gnome.org/show_bug.cgi?id=698981
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
It tries to run glib-compile-schemas and glib-mkenums, which
we won't have in the runtime tree.
Anyways it's kind of a dumb test since the best test for
compilation tools is...compiling things, which we already
do frequently.
Although none of the in-tree GSocketConnectable types need it, other
types (like SoupAddress) may find it useful to be able to pass a URI
and a default-port to GProxyAddressEnumerator separately (the same way
you can with GNetworkAddress). So add a default-port property.
https://bugzilla.gnome.org/show_bug.cgi?id=698877
Not all systems have /usr/bin/true. Some have it in /bin/true.
Instead of trying to guess a hardcoded path to find it, let
g_app_info_create_from_commandline() internally search PATH
to find the program.
https://bugzilla.gnome.org/show_bug.cgi?id=698655
Add support for serialising a GIcon to a GVariant and deserialising the
result back to a GIcon.
This solves a number of problems suffered by the existing to_string()
API, primarily these:
- not forcing the icon to be a utf8 string means that we can
efficiently encode a PNG (ie: just give the array of bytes)
- there is no need to ensure that proper types are loaded before using
the deserialisation interface. 'Foreign' icon types will probably
emit a serialised format the deserialises to a GBytesIcon.
We additionally clearly document what is required for being a consumer
or implementation of #GIcon.
Further patches will be required to GdkPixbuf and GVfsIcon to bring
their implementations in line with the new rules (essentially: introduce
implementations of the new serialize() API).
https://bugzilla.gnome.org/show_bug.cgi?id=688820
Expand and formalise the syntax for detailed action names, adding a
well-documented (and tested) public parser API for them.
Port the only GLib-based user of detailed action names to the new API:
g_menu_item_set_detailed_action(). The users in Gtk+ will also be
ported soon.
https://bugzilla.gnome.org/show_bug.cgi?id=688954
We need to close the stream *before* applying the file modes, because
g_file_replace() allocates a temporary file. At the moment we're
applying the modes to the extant file, then immediately rename()ing
over it with the default perms.
This regressed with commit 166766a89f.
The real fix here is to have g_file_create_with_info() so that we can
atomically create a file with the permissions we want.
https://bugzilla.gnome.org/show_bug.cgi?id=696014
Commit f641699 (for bug 675333) introduced a check whether the Exec= program in
a .desktop actually exists. This broke the /appinfo/mime/* test cases which use
executable names like "my_app".
Use real ones instead (like "echo" and "sleep"), and add a new
/appinfo/mime/ignore-nonexisting test case which verifies that
g_desktop_app_info_new() indeed ignores nonexisting executables.
https://bugzilla.gnome.org/show_bug.cgi?id=695191
Some (broken) toolchains for example trip up
-Werror=missing-prototypes in system headers. This patch allows
people to skip the formerly hardcoded "baseline" warnings.
https://bugzilla.gnome.org/show_bug.cgi?id=694757
If a GNetworkAddress is created with a hostname like "fe80::xxx%em1",
make sure that the scope_id corresponding to "em1" is present in the
GSocketAddresses it returns when used as a GSocketConnectable.
https://bugzilla.gnome.org/show_bug.cgi?id=684404
Add GSimpleProxyResolver, for letting people do static proxy
resolution, and to use as a base class for other resolvers (such as
GProxyResolverGnome).
https://bugzilla.gnome.org/show_bug.cgi?id=691105
Add a proxy-resolver property to GSocketClient, to allow overriding
proxy resolution in situations where you need to force a particular
proxy rather than using the system defaults.
https://bugzilla.gnome.org/show_bug.cgi?id=691105
This is a GLib reimplementation of dbus_address_escape_value().
It's useful if you want to construct a D-Bus address from pieces:
for instance, if you have a listening Unix socket whose path is known,
and you want to connect a D-Bus peer to it.
Bug: https://bugzilla.gnome.org/show_bug.cgi?id=693673
Reviewed-by: Colin Walters <walters@verbum.org>
[amended to add Since: 2.36 as per review]
Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
While compiling, libtool will say that undefined symbols are not allowed, and
will refuse to make you a dll. This is only one line, easy to miss. And it
doesn't prevent `make' from completing successfully.
The code this patch adds is from other Makefile.am files that use
$(no_undefined). It's absence in gio is, most likely, an oversight.
Fixes#692058
The flowinfo and scope_id fields of struct sockaddr_in6 are in host
byte order, but the code previously assumed they were in network byte
order. Fix that.
This is an ABI-breaking change (since before you would have had to use
g_ntohl() and g_htonl() with them to get the correct values, and now
that would give the wrong values), but the previous behavior was
clearly wrong, and no one ever reported it, so it is likely that no
one was actually using it.
https://bugzilla.gnome.org/show_bug.cgi?id=684404
Add read_async() and skip_async() tests to buffered-input-stream.
Fix and re-enable filter-streams's existing close_async() tests, and
add read_async(), skip_async(), and write_async() tests as well. Also,
redo the tests to use dummy GFilterInputStream and GFilterOutputStream
subclasses rather than GBufferedInput/OutputStream, so that we're
testing the base filter stream implementations of everything (since
the buffered stream overrides are already getting tested in the
buffered-input-stream and buffered-output-stream tests anyway).
Add a skip_async() test to unix-streams. (This one would crash without
the bugfix in the previous commit.)
We only want to control the default visibility for our five main
installable libraries: libglib, libgthread, libgmodule, libgobject,
libgio. We should therefore only set -fvisibility=hidden when building
those.
Use a separate substitution variable for this purpose.
Using CFLAGS directly leads to some modules built in testcases not
exporting their symbols (and then the tests fail). It also affects the
fam file monitoring module.
Colin had originally done it this way in his visibility patch series but
I failed to understand why so I didn't copy it. Now I do.
Also: revert changes made to two testcases in an attempt to work around
this issue.
https://bugzilla.gnome.org/show_bug.cgi?id=691756
One of our testcases builds a small giomodule for testing the loading of
modules containing resources. Unfortunately, this module gets built
using the same CFLAGS as the rest of GLib, including the visibility
flags (defaulting to hidden).
Use "config.h" to get a declaration of _GLIB_EXTERN that will export
symbols properly and use it to annotate the necessary APIs.
Some OS (e.g. OpenBSD) do not implement IP v4-mapped addresses. When
this is the case, then we get a "Connection refused", so force the test
to pass to that further tests can run.
https://bugzilla.gnome.org/show_bug.cgi?id=686058
a5876e5f made GMemoryInputStream subclassable, but accidentally broke
read_async() and skip_async() in the process. The immediately
following e7983495 fixed read_async() (and added a test for it), but
skip_async() accidentally got... skipped.
Fix it now and add a test for it.
Also, GMemoryInputStream's skip_async() was assuming that skip() could
never fail, which is true of its own implementation, but might not be
true of a subclass's, so do proper GError handling too.
This will let us drop the dbus-python dependency.
The C version does not 100% reproduce all the hash table
and array manipulation of the python version, but the tests
do not rely on it anyway.
This greatly simplifies the test since everything is now in a single
process and possible bugs / quirks in libdbus-1 will not interfere
with the tests. On the other hand, we no longer test interoperability
with libdbus-1. This is somewhat moot, however, since other tests that
involve a message bus (e.g. GTestDBus users which include most of the
GDBus test suite itself) will test this.
Also ensure that we don't pollute existing D-Bus keyrings for the
DBUS_COOKIE_SHA1 authentication method (e.g. files in the
~/.dbus-keyrings directory) by setting the environment variables
G_DBUS_COOKIE_SHA1_KEYRING_DIR and
G_DBUS_COOKIE_SHA1_KEYRING_DIR_IGNORE_PERMISSION.
All in all, this change avoids some thorny issues where the GDBus and
libdbus-1 implementations disagree on whether an item in the D-Bus
keyring is still valid (items have an age etc.). In reality, since the
DBUS_COOKIE_SHA1 authentication method is never used in production,
this is never hit in production. This bug was, however, frequently hit
if you just ran the test suite repeatedly for 15 minutes or so.
Also add TODO items to mention that we currently don't test corner
cases involving
- DBUS_COOKIE_SHA1 timeouts
- libdbus-1 interoperability
Signed-off-by: David Zeuthen <zeuthen@gmail.com>
If tasks block waiting for other tasks to complete then the system can
end up starved for threads. Avoid this by bumping up max-threads in
that case.
This also reverts 7b1f8c58 and reverts max-threads for GTask's
GThreadPool back to 10.
https://bugzilla.gnome.org/show_bug.cgi?id=687223
This is a new convenience method designed to simplify some use
cases of GFileEnumerator, by making it easy to get the next file
from a file enumerator.
https://bugzilla.gnome.org/show_bug.cgi?id=690083
Install a public "gnetworking.h" header that can be used to include
the relevant OS-dependent networking headers. This does not really
abstract away unix-vs-windows however; error codes, in particular,
are incompatible.
gnetworkingprivate.h now contains just a few internal URI-related
functions
Also add a g_networking_init() function to gnetworking.h, which can be
used to explicitly initialize OS-level networking, rather than having
that happen as a side-effect of registering GInetAddress.
https://bugzilla.gnome.org/show_bug.cgi?id=623187
The three processes this test creates need to be executed
in order, and g_usleep was used to guarantee that.
However, under heavy load, that is not enough. Instead,
wait until the children start by making sure they have
written to stdout before proceeding any further.
https://bugzilla.gnome.org/show_bug.cgi?id=664627
Really, the memory output stream API is too warped around the model
where it's a fixed size buffer that you've already allocated. Even in
C, I find myself always wanting to use it to just accumulate data into
an arbitrary-sized buffer it allocates.
Unfortunately, it's also not usable from bindings because it's not
common to bind g_free() and g_realloc(), but if you just pass NULL, you
get the default of a fixed size, which is useless as per above.
I am going to use this from a gjs test case, and the GSubprocess test
cases also will use it.
https://bugzilla.gnome.org/show_bug.cgi?id=688931
Darwin's poll doesn't change revents if there are no available events, though it returns 0. Initialize the fd.revents to 0 so that the test passes.
That reveals a test failure, though, because with socket streams it takes time for an event to pass through the socket. Provide an 80-usec delay to allow time for the propagation.
Allow GDBusObjectManagerClient to work on peer to peer DBus
connections. Don't require that a unique bus name is available
for the object manager, if the owned bus name is NULL.
https://bugzilla.gnome.org/show_bug.cgi?id=686920
Sometimes the poll duration in the /socket/timed_wait test is slightly
bigger than the requested 100000, causing failures like:
GLib-GIO:ERROR:socket.c:620:test_timed_wait:
assertion failed (poll_duration < 110000): (110057 < 110000)
Adjust the test to allow some jitter in the "too high" direction.
https://bugzilla.gnome.org/show_bug.cgi?id=686458
Very many testcases, some GLib tools (resource compiler, etc) and
GApplication were calling g_type_init().
Remove those uses, as they are no longer required.
https://bugzilla.gnome.org/show_bug.cgi?id=686161
GTask is a replacement for GSimpleAsyncResult and GIOScheduler, that
also allows for making cancellable wrappers around non-cancellable
functions (as in GThreadedResolver).
https://bugzilla.gnome.org/show_bug.cgi?id=661767
This is the expected (and sane) behavior - without this bug-fix you'd
have to add "Since" to every member of a newly added D-Bus interface.
Also show-case this in the codegen example.
Signed-off-by: David Zeuthen <zeuthen@gmail.com>
Adding the --sourcedir option fixes these:
/path/to/src/gio/tests/test2.gresource.xml: Error on line 5 char 1: Failed to locate 'test1.txt' in current directory.
/path/to/src/gio/tests/test3.gresource.xml: Error on line 5 char 1: Failed to locate 'test1.txt' in current directory.
/path/to/src/gio/tests/test4.gresource.xml: Error on line 5 char 1: Failed to locate 'test1.txt' in current directory.
/path/to/src/gio/tests/test.gresource.xml: Error on line 5 char 1: Failed to locate 'test1.txt' in current directory.
The test was assuming that all cancelled ops would finish within a
certain amount of time, but this often failed under valgrind. Instead,
just run the loop until all of the ops have actually finished.
https://bugzilla.gnome.org/show_bug.cgi?id=682560
On slower platforms, the overhead of the 240 D-BUS Sleep calls is larger than
the current maximum of 6 seconds. A run on a Panda board sometimes fails with
ERROR:/build/buildd/glib2.0-2.33.8/./gio/tests/gdbus-threading.c:409:test_method_calls_on_proxy:
assertion failed (elapsed_msec < 6000): (7365 < 6000)
Bump maximum time to 8 seconds to be more resilient to this.
https://bugzilla.gnome.org/show_bug.cgi?id=682222
Because it now handles EINTR. And we should do so. While most people
use Linux, which tries very hard to avoid propagating EINTR back up
into userspace, it can still happen.
https://bugzilla.gnome.org/show_bug.cgi?id=682833
The async test had several problems:
- It created a proxy and did not launch a main loop, meaning that its
callback would usually not get called, or, if it did get called, the
test harness would have taken down the connection already, causing an
assertion failure when the proxy had an error.
- It was dependent on the proxy test to set up the server and would fail
because some properties were modified by that test.
https://bugzilla.gnome.org/show_bug.cgi?id=674805
* In order to add contstruct properties to an abstract base
calls, and retain ABI stability, the base class must add a
default implementation of those properties.
* We cannot add a default implementation of certificate-bytes
or private-key-bytes since certificate and private-key properties
are writable on construct-only.
This reverts commit 541c985869.
https://bugzilla.gnome.org/show_bug.cgi?id=682081
Implement test case suggested by Ryan Lortie on bug:
https://bugzilla.gnome.org/show_bug.cgi?id=679288
"There is a potential race here that's really unlikely to happen, but
here we go: We are trying to read from the same socket in two threads.
Some data comes. That causes the poll() in both threads (above) to
finish running. Then the cancellable is checked above. We now find
ourselves here. Only one thread will read the data. The other will
block on this function. Then the user may cancel the cancellable while
we are blocked here, but we will stay blocked...."
test_create_delete() assumes that if it creates a file and then
immediately deletes it, that the file monitor will notice this and
record it as a create followed by a delete. But that won't work with
GPollFileMonitor, which will just think nothing changed. So skip the
test in that case.
https://bugzilla.gnome.org/show_bug.cgi?id=669331
This looks like it was stubbed out but not implemented; the vtable
entry dates to commit 3781343738 which
is just alex's initial merge of gio into glib.
I was working on some code that wants an asynchronous rm -rf
equivalent, and so yeah, this is desirable.
https://bugzilla.gnome.org/show_bug.cgi?id=680760
Add a test that the decompressor input streams handle truncated data
correctly. (They do; I wrote the test thinking there was a bug there,
but there isn't.)
Also, rename the "corruption" tests to "roundtrip", since "corruption"
makes it sound like we're testing how the converters deal with
corrupted data, as opposed to merely testing that they don't corrupt
data themselves. And fix the bug reference.
Many (if not "almost all") programs that spawn other programs via
g_spawn_sync() or the like simply want to check whether or not the
child exited successfully, but doing so requires use of
platform-specific functionality and there's actually a fair amount of
boilerplate involved.
This new API will help drain a *lot* of mostly duplicated code in
GNOME, from gnome-session to gdm. And we can see that some bits even
inside GLib were doing it wrong; for example checking the exit status
on Unix, but ignoring it on Windows.
https://bugzilla.gnome.org/show_bug.cgi?id=679691
For a D-Bus property with name "Type" (fairly common), we used to
generate a GObject property with name "type-" and C accessors
get_type_() (to avoid clashing with the GType getter), set_type_()
(for symmetri).
However, the rules for GObject property names are fairly rigid and
specifically prohibit names ending in a dash.
Therefore change things so the chosen GObject property name is "type"
but preserve the naming rules for the C getter and setter (for the
same reasons: avoiding name clashing and symmetri).
This change does break the API of generated code (but only on the
GObject property level, the C symbols are not changed) but strictly
speaking the behavior was undefined since "type-" was an invalid
GObject property name.
Also add a test case for this.
Bug 679473.
https://bugzilla.gnome.org/show_bug.cgi?id=679473
Signed-off-by: David Zeuthen <zeuthen@gmail.com>
After fixing bug 674452 this test case now reliably fails, as "ABC abc" is text
and definitively not PowerPoint. It previously worked as g_content_type_guess()
was reading beyond the boundary of the data due to specifying -1 as data
length.
Update that test case to expect a PO template instead, and add two more with a
definitive PO template syntax and some binary data. We do not currently have a
MIME magic for PowerPoint, so we cannot actually detect it with certainty, but
at least make sure that the returned MIME type is correct.
https://bugzilla.gnome.org/show_bug.cgi?id=678941
g_content_type_guess() requires specifying a valid data length. Fixes a
segfault when running the test.
Also add an explicit check for this and return XDG_MIME_TYPE_UNKNOWN when
data_size is specified as -1, to avoid crashing.
https://bugzilla.gnome.org/show_bug.cgi?id=674452
Sometimes the poll duration in the /socket/timed_wait test is slightly lower
than the requested 100000, causing failures like
ERROR:/build/buildd/glib2.0-2.33.2/./gio/tests/socket.c:619:test_timed_wait:
assertion failed (poll_duration > = 100000): (99240 >= 100000)
FAIL
Adjust the test to also allow some jitter in the "too small" direction, similar
to the already existing span for "slightly too large".
https://bugzilla.gnome.org/show_bug.cgi?id=678881
make sure the proxy threads are in the "waiting for a connection"
state when we do the final cleanup, or else there are race conditions
involving which thread processes the GCancellable cancellation first.
