db: add indices where queries takes more than 100 ms

CREATE INDEX cc_pull_request_id_idx ON changeset_comments USING btree (pull_request_id);
CREATE INDEX cs_changeset_comment_id_idx ON changeset_statuses USING btree (changeset_comment_id);
CREATE INDEX cs_pull_request_id_idx ON changeset_statuses USING btree (pull_request_id);
CREATE INDEX pr_org_repo_id_idx ON pull_requests USING btree (org_repo_id);
CREATE INDEX pr_other_repo_id_idx ON pull_requests USING btree (other_repo_id);

comments: introduce lazy join of ChangesetStatus on ChangesetComment

99% of all uses of ChangesetComment will also need the optional corresponding
ChangesetStatus (if any) on the status_change relationship. Fetching it on
demand gives a lot of roundtrips and might be slow ... but adding explicit bulk
queries everywhere do not seem feasible. Adding lazy=joined in the data model
seems like the best solution.

Loading a pull request with 10 comments on a slow repo goes from 3.5 s to
2.7 s.

comments: optimize notification creation

.append to the sqlalchemy collection is apparently expensive. Setting the
reference the other way around is much faster. Adding a notification for a
single user goes from 0.25 s to 0.001 s.

On a slow repo and a pullrequest with 14 reviewers, adding a comment goes from
8.5 s to 0.8 s.

changeset status: drop default 'unreviewed' status for changesets on PRs

The default is expensive to compute and doesn't scale - especially not with the
current data model where all pull-request revisions are stored in a single
field and not just can be indexed and join in the database.

cleanup: avoid some 'except Exception' catching - catch specific exceptions or log it and show what happened

This has a risk of introducing regressions ... but we want to get rid of all
exception muting and make the whole system less fragile and easier to debug.

helpers: refactor urlify_commit, extract urlify_issues function

A slight improvement of code, but also prepares for other wip changes.

db: just fail badly if a repo can't be loaded (probably missing in file systems)

The error condition could perhaps be reported to the user in a better way, but
propagating the VCS error seems like a better approximation of that than just
logging it and failing silent.

cache invalidation: don't commit session if nothing changed

It seems like a good idea to avoid unnecessary commits, considering we use
expire_on_commit=True.

It would perhaps be even better to postpone the commit to the end of the
session - that is for another day.

pullrequests: drop expensive listing of unrelated changesets that cannot be used for update

It rarely gave any relevant info and could take several seconds to compute.

Instead, just warn if the branch name has another head.

Loading a pullrequest page on a slow repo did in one setup go from 7 s to 3.5 s.

hg: reimplement branch listings more efficiently

Gives a 10 x speedup - which is noticable on big slow repos where the old
implementation took several seconds.

pullrequests: store the 'a' branch name in template context

This is not used upstream yet but is useful for customizations. Getting this
change in early avoids conflicts when developing other changes.

hg: drop support for hg web.hidden

Kallithea is not hgweb and should not try to support hgweb config options.

If someone wants to hide repos, they should just remove all access to them.

(_If_ Kallithea should support some kind of more thorough repo hiding, it
should be implemented differently: The hidden repos should not be in the
database and should thus instead be excluded when scanning for repos.)

styling: don't loop on loading kallithea-logo.png after kallithea-logo.svg failed

Bad configuration of static files or static_files=false could make the page
loop when trying to apply the svg/png fallback.

Instead, let the error handler remove itself when it is used the first time.

error: remove allmost unused code for redirect on error page

It is much better to just stop and show the error message.

json: drop simplejson fallback and just use stdlib

The performance gain from using simplejson is questionable and undocumented for
our use. Keep it simple.

pullrequests: show pending reviewer changes again - this time with "(not saved)"

b5795554c2ca lost some of the changes from 00c76c4e019c.

user groups: drop add/remove all buttons

The buttons can easily mess the old selection up and do not add much value
compared to multi select.

paster: fix update_repoinfo command

Less bitrot.

javascripts: replace YUI with jquery in kallithea/templates/pullrequests/pullrequest_show.html

This patch removes the '.show-inline-comments' event handler.
'.show-inline-comments' does not exist in diff_block.diff_block_simple,
it exists in diff_block.diff_block.

diff: don't split lines on bare CR as python splitlines do

The line count from the @@ lines and the actual count would get out of sync and
an incomplete diff would be shown.

This is slower than native splitlines ... but that is what it takes to handle
it correctly. (Except that we perhaps could use .split('\n') and patch the
result ...)

diff: eat trailing tabs in ---/+++ lines of diffs

There is usually no date in the diffs we parse but there might still be the tab
that was used as separator between filename and date.

The a_file and b_file entries are usually not used ... but nice to have them
right.

diff: fix diff of renamed files with whitespace in their names

Before, header parsing would fail.

diff: preserve leading space on empty lines - don't match in re if we don't plan to replace

Fix invisible bug from 7c094db329b3, exposed by future exact parsing of diffs.

Translated using Weblate (Russian)

Currently translated at 78.9% (850 of 1077 strings)

Translated using Weblate (German)

Currently translated at 26.5% (286 of 1077 strings)

Translated using Weblate (German)

Currently translated at 23.2% (250 of 1077 strings)