| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
|
|
|
|
|
|
|
|
|
| |
This fixes crashes in gnome-software when concurrently calling
as_app_get_unique_id() from multiple threads.
https://bugzilla.redhat.com/show_bug.cgi?id=1655775
https://bugzilla.redhat.com/show_bug.cgi?id=1643702
https://bugzilla.redhat.com/show_bug.cgi?id=1640446
https://bugzilla.redhat.com/show_bug.cgi?id=1530719
|
|
|
|
|
|
|
|
|
| |
This fixes, among other things, appdata validation when the version goes
from 9.5 to 10, or 0.9 to 1.
This partially fixes https://github.com/hughsie/appstream-glib/issues/270
We're still failing to correctly compare "Build 9.5" to "Build 10".
|
|
|
|
| |
https://github.com/hughsie/appstream-glib/issues/270
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Keep a ref on the AsApp in the worker thread to make sure it doesn't get
destroyed in another thread while being processed.
https://bugzilla.redhat.com/show_bug.cgi?id=1662130
https://bugzilla.redhat.com/show_bug.cgi?id=1661701
https://bugzilla.redhat.com/show_bug.cgi?id=1661382
https://bugzilla.redhat.com/show_bug.cgi?id=1658140
https://bugzilla.redhat.com/show_bug.cgi?id=1644026
https://bugzilla.redhat.com/show_bug.cgi?id=1639558
https://bugzilla.redhat.com/show_bug.cgi?id=1570096
|
| |
|
| |
|
|
|
|
|
| |
When returning early we were leaking GPtrArray *probs which never got
freed. Change it to use g_autoptr to avoid that.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This adds fine-grained locking around priv->array and various
priv-stored hash table use. There are more thread safe issues in
AsStore, but this should take care of those that cause frequent
gnome-software crashes in F29.
This may regress the perfomance a bit because it changes a few places to
keep a deep copy to simplify locking, but I haven't observed any visible
performance regressions in gnome-software. Also, gnome-software is in
the process of switching to libxmlb anyway so it shouldn't matter a
whole lot in the long run.
This patch takes care to make sure that locking is fine grained enough
so that we can be sure it doesn't lead into deadlocks, and also makes
sure that we never invoke any callbacks (signals) while locked to
prevent deadlocks when a client app calls back into AsStore code.
|
|
|
|
|
|
|
|
|
|
| |
This is strictly not necessary for making gnome-software's AsStore use
thread safe as gnome-software doesn't actually use this function, but
let's fix it anyway while I'm at it.
This also updates tests that test the performance of the function as the
deep copying makes it noticably slower (but that's fine because nothing
actually uses it and it's still reasonably fast).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
gnome-software spawns a new worker thread for each of its plugin
operations and occasionally this leads to issues where one thread is
reading from AsStore, and another one is changing it. There's also file
monitor callbacks in AsStore that run in a yet another thread and can
update internal data structures as well.
This leads to a situation where functions returning pointers to internal
data structures can't guarantee the data structure lifetime because
another thread might be changing it in the background.
To fix this, this commit adds new as_store_dup_apps() and
as_store_dup_apps_by_id_merge() functions that return a deep copy of the
internal structure, and updates existing "transfer container"
as_store_get_apps_by_id() to return a deep copy, instead just returning
a reffed container.
|
| |
|
| |
|
|
|
|
|
| |
This allows the search to match on e.g. "eog" for an app that has
the ID of "org.gnome.org".
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, as_store_from_file() would either return
G_IO_ERROR_NOT_FOUND or AS_STORE_ERROR_FAILED if the given file did not
exist, depending on its file extension. That is not particularly helpful
for callers.
Instead, guarantee to always return G_IO_ERROR_NOT_FOUND if the file was
not found.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
|
|
|
|
|
|
|
|
|
|
| |
This allows the set of rating IDs in the AsContentRating to be returned,
allowing the caller to iterate over them to check their values, rather
than having to hard-code their own list of OARS rating IDs.
Includes a basic test.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
|
|
|
|
| |
Also tweak a few bits of punctuation I noticed at the same time.
|
| |
|
| |
|
|
|
|
| |
Sometimes we don't want to do the firmware-style heuristics.
|
|
|
|
| |
This allows us to drop some hacky code in GNOME Software.
|
|
|
|
| |
This allows us to load in a file and export it as AppData XML.
|
|
|
|
| |
This allows us to contruct a AsContentRating without using XML.
|
|
|
|
| |
See https://blogs.gnome.org/hughsie/2017/07/05/using-a-reverse-style-application-ids-in-your-application/
|
| |
|
|
|
|
| |
If this needs limiting it needs to be done on the generator.
|
| |
|
|
|
|
|
|
|
| |
This quirk allows apps to marked as having an additional level of checking
performed on the developer. This is to give confidence to the user that this
is the official version of an app. This is a common design pattern used in
modern packaging systems (e.g. Apple App Store, Google Play).
|
|
|
|
|
|
| |
The existing functions may block, which isn't ideal in IO-sensitive
contexts. To address this, add async variants that run the functions
in a thread.
|
| |
|
| |
|
|
|
|
|
|
|
| |
This allows upstream projects to prepare (and translate) the release data
before know the actual release date.
Fixes: https://github.com/hughsie/appstream-glib/issues/246
|
| |
|
| |
|
|
|
|
| |
Firmware should now be distributed using the LVFS and consumed by fwupd.
|
|
|
|
| |
Spotted by Coverity, many thanks.
|
|
|
|
| |
Spotted by Coverity, many thanks.
|
|
|
|
|
|
|
| |
Two problems here:
* Don't force an unknown icon to CACHED when reading from a desktop file
* Don't try to convert an icon to a CACHED icon when it's already CACHED.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
This allows us to trivially dedupe kudos and URL kinds.
|
| |
|
| |
|
| |
|