Monty Taylor [Thu, 7 Feb 2019 02:25:23 +0000 (20:25 -0600)]
Support n as a line highlight prefix (#5987)
We're working on rolling out gitea as a replacement for a fairly
large cgit installation. Part of this involves a metric ton of
mod_rewrite rules so that links people have out in the wild will
still work. Unfortunately, cgit uses #n1 in the url to indicate
a single line highlight, which doesn't get passed to the server
and therefore can't be rewritten.
We've got a local workaround by shoving a script block into a
custom header template, but thought it might be nicer to upstream
a simple patch to support #n as a prefix for single line highlights.
(luckily for simplicity, cgit does not support ranges)
Signed-off-by: Monty Taylor <mordred@inaugust.com>
silverwind [Wed, 6 Feb 2019 07:09:13 +0000 (08:09 +0100)]
UI: Dashboard tweaks (#5974)
- Don't show divider below heatmap when it's hidden via media-query
- Make divider between user menu adhere to container width
- Remove excessiv margins on heatmap
silverwind [Tue, 5 Feb 2019 21:59:26 +0000 (22:59 +0100)]
UI: Repo header tweaks (#5945)
* UI: Repo header tweaks
- Use basic buttons on the header buttons, matching gogs
- Make 'Manage topic' text smaller, remove margin-left with no topics
present
- Move various inline styles to CSS
- Use flexbox on header title and buttons
zeripath [Sun, 3 Feb 2019 23:56:53 +0000 (23:56 +0000)]
Fix ssh deploy and user key constraints (#1357) (#5939)
1. A key can either be an ssh user key or a deploy key. It cannot be both.
2. If a key is a user key - it can only be associated with one user.
3. If a key is a deploy key - it can be used in multiple repositories and the permissions it has on those repositories can be different.
4. If a repository is deleted, its deploy keys must be deleted too.
We currently don't enforce any of this and multiple repositories access with different permissions doesn't work at all. This PR enforces the following constraints:
- [x] You should not be able to add the same user key as another user
- [x] You should not be able to add a ssh user key which is being used as a deploy key
- [x] You should not be able to add a ssh deploy key which is being used as a user key
- [x] If you add an ssh deploy key to another repository you should be able to use it in different modes without losing the ability to use it in the other mode.
- [x] If you delete a repository you must delete all its deploy keys.
zeripath [Sun, 3 Feb 2019 14:00:48 +0000 (14:00 +0000)]
Fix notifications on pushing with deploy keys by setting hook environment variables (#5935)
The gitea prerecieve and postrecieve hooks and the gitea PushUpdate function require that the PusherID and PusherName are real users. Previously, these environment variables were not being set when using a deploy key - the main result being that pushing to empty repositories meant that is_empty status was not changed.
I've also added an integration test to ensure that the is_empty status is updated on pushing with a deploy key.
There is a slight issue in that the deploy key is now considered a proxy for the owner - we don't have a way of separating out the deploy key from the owner at present. This can be fixed in another PR.
zeripath [Sun, 3 Feb 2019 11:19:28 +0000 (11:19 +0000)]
Fix #5866: Silence console logger in gitea serv (#5887)
By default, if `setting.NewContext()` prints out any warning logs, these are printed to the stdout breaking `git receive-pack` etc. meaning that even if there is a warning because of a minor problem in your app.ini but gitea starts despite this - you **CANNOT** push or pull over SSH.
This PR disables the console logger whilst in `serv.go`
silverwind [Sat, 2 Feb 2019 16:58:28 +0000 (17:58 +0100)]
Fix wrapping long code lines in UI (#5927)
Wrapping was initially added in https://github.com/go-gitea/gitea/pull/2789
but it is currently disabled because the rule that applies wrapping to
the wrap class did not have enough CSS specificity.
Resolved the issue by using a general selector that matches all code
boxes. The previous wrap class was removed because it had only been
inconsistenly applied to various code boxes and because I found no easy
way to add classes to code boxes rendered in Markdown.
Also removed a seemingly useless :before on code view and added padding
to restore the spacing.
zeripath [Wed, 30 Jan 2019 22:17:43 +0000 (22:17 +0000)]
Recover panic in orgmode.Render if bad orgfile (#4982) (#5903)
This PR protects against the panic referred to in chaseadmsio/goorgeous#82
by recovering from the panic and just returning the raw bytes if
there is an error.
zeripath [Wed, 30 Jan 2019 22:00:00 +0000 (22:00 +0000)]
Provide better panic handling (#5902)
This PR gitea'ises the macaron.Recovery() handler meaning that in
the event of panic we get proper gitea 500 pages and the stacktrace
is logged with the gitea logger.
zeripath [Thu, 24 Jan 2019 14:12:17 +0000 (14:12 +0000)]
Ensure valid git author names passed in signatures (#5774)
* Ensure valid git author names passed in signatures
Fix #5772 - Git author names are not allowed to include `\n` `<` or `>` and
must not be empty. Ensure that the name passed in a signature is valid.
* Account for pathologically named external users
LDAP and the like usernames are not checked in the same way that users who signup are.
Therefore just ensure that user names are also git safe and if totally pathological -
Set them to "user-$UID"
* Add Tests and adjust test users
Make our testcases a little more pathological so that we be sure that integration
tests have a chance to spot these cases.
zeripath [Wed, 23 Jan 2019 08:56:51 +0000 (08:56 +0000)]
Pooled and buffered gzip implementation (#5722)
* Pooled and buffered gzip implementation
* Add test for gzip
* Add integration test
* Ensure lfs check within transaction
The previous code made it possible for a race condition to occur whereby a LFSMetaObject could be checked into the database twice. We should check if the LFSMetaObject is within the database and insert it if not in one transaction.
* Try to avoid primary key problem in postgres
The integration tests are being affected by
https://github.com/go-testfixtures/testfixtures/issues/39 if we set the
primary key high enough, keep a count of this and remove at the end of
each test we shouldn't be affected by this.
Paul Brackin [Sun, 20 Jan 2019 09:30:37 +0000 (01:30 -0800)]
Ensure error is logged on db reconnection attempts (#5780)
Unfortunately a scoping problem was missed on the previous PR which meant that the only nil error was logged on failure to connect to the db. This pr fixes this.
zeripath [Sun, 20 Jan 2019 01:20:52 +0000 (01:20 +0000)]
Fix the v78 migration script (#5776)
Unfortunately the last fix didn't completely fix the migration to v79 of the db
due to bug with schema locking during Sync2. This should fix this issue.