* Fix ParsePatch to work properly with quoted diff --git string
Currently ParsePatch fails when a diff contains a quoted diff line like:
diff --git "a/file" "b/file"
This patch makes it properly parse the line when that happens.
Fixes #6309
* Add test for regular case while here
* Simplify string modification
* Use url.PathEscape to escape the branchname
* GetRepositoryByOwnerAndName should also have url.PathEscape as the owner and reponame are provided by the client
* Replace linkRegex with xurls library
Rather than maintaining a complicated regex to match URLs for
autolinking, gitea can use this existing go library that takes care of
the matching with very little code change to gitea itself. After
spending a while trying to find the perfect regex for all cases this library
still works better as it is more flexible than a single regex ever will be.
This will also fix the following issues: #5844#3095#3381
This passes all our current tests and I've added new ones mentioned in
those issues as well.
* Use xurls.StrictMatchingScheme instead of xurls.Strict
This is much faster and we only care about https? links to preserve
existing behavior.
The visitLinksForShortLinks feature would look inside of an <a> tag and
run shortLinkProcessorFull on any text, which attempts to create links
out of potential 'short links' like [[test]] [[link|example]] etc...
This makes no sense because you can't have nested links within an <a>
tag. Specifically, the html5 standard says <a> tags can't include
interactive content if they contain the href attribute:
http://w3c.github.io/html/single-page.html#the-a-element
And also defines an <a> element with a href attribute as interactive:
http://w3c.github.io/html/single-page.html#interactive-content
Therefore you can't really put a link inside of another link. In
practice none of this works anyways since browsers won't render it, it
would probably be broken if they tried, and it is causing a bug
(#4946). No current tests rely on this behavior either.
This removes the feature and also explicitly excludes the
current visitNodeForShortLinks from looking in <a> tags.
In #6211, we started creating repo_redirects for ownership transfers,
however that opens an edge case where a user might perform the
following sequence:
rename org1/repo1 -> org1/repo2 (creates org1/repo1 redirect)
transfer org2/repo1 -> org1/repo1 (org1/repo1 redirect continues to exist)
rename org1/repo1 -> org1/repo3 (fails due to existing org1/repo1 redirect)
This change ensures that each time we rename or transfer a repo,
we delete any existing redirects at the target location. This
already happens when a new repo is created. By doing this we ensure
that we'll never have both a repo and a redirect at the same location.
Signed-off-by: James E. Blair <jeblair@redhat.com>
