When fetching remotes for conflict checking, skip unnecessary and
potentially slow writing of commit graphs.
In a test with the Blender repository, this reduces conflict checking
time for one pull request from about 2s to 0.1s.
When a change is pushed to the default branch and many pull requests are
open for that branch, conflict checking can take some time.
Previously it would go from oldest to newest pull request. Now
prioritize pull requests that are likely being actively worked on or
prepared for merging.
This only changes the order within one push to one repository, but the
change is trivial and can already be quite helpful for smaller Gitea
instances where a few repositories have most pull requests. A global
order would require deeper changes to queues.
There have been a number of reports of PRs being blocked whilst being
checked which have been difficult to debug. In investigating #23050 I
have realised that whilst the Warn there is somewhat of a miscall there
was a real bug in the way that the LevelUniqueQueue was being restored
on start-up of the PersistableChannelUniqueQueue.
Next there is a conflict in the setting of the internal leveldb queue
name - This wasn't being set so it was being overridden by other unique
queues.
This PR fixes these bugs and adds a testcase.
Thanks to @brechtvl for noticing the second issue.
Fix #23050
and others
---------
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: techknowlogick <techknowlogick@gitea.io>
This arose out of #22451; it seems we are checking using non-global
settings to see if a config value is set, in order to decide whether to
call another global(-indeed) configuration command. This PR changes it
so that both the check and the set are for global configuration.
Close #23137
The old code is too old (8-9 years ago)
Let's try to execute the git commands from git bin home directly.
The verb has been checked above, it could only be:
* git-upload-pack
* git-upload-archive
* git-receive-pack
* git-lfs-authenticate
Before this change, the `app.ini` would get overwritten on each
workspace start, confusing Gitea. It asked for reinstallation each time.
This makes sure the file is written only once by checking it does not
exist before creating it.
---
[Review without whitespace
diff](https://github.com/go-gitea/gitea/pull/23192/files?w=1)
---------
Co-authored-by: delvh <dev.lh@web.de>
The name of the job or step comes from the workflow file, while the name
of the runner comes from its registration. If the strings used for these
names are too long, they could cause db issues.
Close #10468
Without SimpleMDE/EasyMDE, using Simple Textarea, the button text could
be changed when content changes.
After introducing SimpleMDE/EasyMDE, there is no code for updating the
button text.
GetActiveStopwatch & HasUserStopwatch is a hot piece of code that is
repeatedly called and on examination of the cpu profile for TestGit it
represents 0.44 seconds of CPU time. This PR reduces this time to 80ms.
---------
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: KN4CK3R <admin@oldschoolhack.me>
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: delvh <leon@kske.dev>
Using `touchstart` for `click` events is a black magic for mobile
browsers (Google: `fastclick`).
However, it causes many UX problems if the fastclick is used without
careful design.
Fomantic UI uses this fastclick for its `dimmer` and `dropdown`, it
makes mobile users feel strange when they "touch" the dropdown menu.
This PR uses a simple patch to fix that behavior. Then the Fomantic
dropdown only uses `click` for click events.
This PR is simple enough and won't cause hidden bugs even if the patch
doesn't work. In the future, if there are more patches for Fomantic UI,
the patches could be placed in a directory like
`web_src/fomantic/patches/001-fix-click-touchstart`, etc.
![image](https://user-images.githubusercontent.com/2114189/220551915-bd28e8cc-507f-43c7-bb4a-b24f7ff3934d.png)
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
This includes pull requests that you approved, requested changes or
commented on. Currently such pull requests are not visible in any of the
filters on /pulls, while they may need further action like merging, or
prodding the author or reviewers.
Especially when working with a large team on a repository it's helpful
to get a full overview of pull requests that may need your attention,
without having to sift through the complete list.
Unfortunately xorm's `builder.Select(...).From(...)` does not escape the
table names. This is mostly not a problem but is a problem with the
`user` table.
This PR simply escapes the user table. No other uses of `From("user")`
where found in the codebase so I think this should be all that is
needed.
Fix #23064
Signed-off-by: Andrew Thornton <art27@cantab.net>
This PR does a bulk update of a lot of our go deps.
I have not included nektos/act and xorm for the following reasons:
* Xorm updates can sometimes be complex and I'd rather do that in a
separate PR
* I think people more update with the actions code should double check
that the latest nektos/act library works correctly.
---------
Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: techknowlogick <techknowlogick@gitea.io>
This notably brings support for GOARCH=loong64, among other fixes.
---------
Signed-off-by: WANG Xuerui <xen0n@gentoo.org>
Co-authored-by: zeripath <art27@cantab.net>
Partially fix #23050
After #22294 merged, it always has a warning log like `cannot get
context cache` when starting up. This should not affect any real life
but it's annoying. This PR will fix the problem. That means when
starting up, getting the system settings will not try from the cache but
will read from the database directly.
---------
Co-authored-by: Lauris BH <lauris@nix.lv>
I found `FindAndCount` which can `Find` and `Count` in the same time
Maybe it is better to use it in `FindProjects`
---------
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
As the title. Label/assignee share the same code.
* Close #22607
* Close #20727
Also:
* partially fix for #21742, now the comment reaction and menu work with
keyboard.
* partially fix for #17705, in most cases the comment won't be lost.
* partially fix for #21539
* partially fix for #20347
* partially fix for #7329
### The `Enter` support
Before, if user presses Enter, the dropdown just disappears and nothing
happens or the window reloads.
After, Enter can be used to select/deselect labels, and press Esc to
hide the dropdown to update the labels (still no way to cancel ....
maybe you can do a Cmd+R or F5 to refresh the window to discard the
changes .....)
This is only a quick patch, the UX is still not perfect, but it's much
better than before.
### The `confirm` before reloading
And more fixes for the `reload` problem, the new behaviors:
* If nothing changes (just show/hide the dropdown), then the page won't
be reloaded.
* If there are draft comments, show a confirm dialog before reloading,
to avoid losing comments.
That's the best effect can be done at the moment, unless completely
refactor these dropdown related code.
Screenshot of the confirm dialog:
<details>
![image](https://user-images.githubusercontent.com/2114189/220538288-e2da8459-6a4e-43cb-8596-74057f8a03a2.png)
</details>
---------
Co-authored-by: Brecht Van Lommel <brecht@blender.org>
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Fix: #22990
---
Before, the return value of the api is always false,regrardless of
whether the entry of `sync_on_commit` is true or false.
I have confirmed that the value of `sync_on_commit` dropped into the
database is correct.
So, I think it is enough to make some small changes.
This PR is a possible solution for issue #22866. Main change is to add a
`author-wrapper` class around author name, like the wrapper added to
message. The `max-width` is set to 200px on PC, and 100px on mobile
device for now.