Previously engine used raw database rows for performance boost. The
benefits were negligibly small, therefore it was changed so that it
returns full entities again. That way serializing job return values
for HTTP API should be easier in the future.
Right now there's a lot of messy code in controllers. Furthermore, there
is no way to interact with szurubooru via vanilla HTTP, since API is
next to non-existent. So, basing upon my experiences from another
project, I plan to:
- Create actual API. It is going to consist of well-defined "jobs" that
do things currently done by controllers. Benefits of such approach are
as follows:
- defining them in their own classes allows to clean up code a lot,
- it allows to abstract from input method (POST data, part of URL,
whatever), and leave processing of these to controllers,
- it allows to make proxy controller, whose purpose would be to let
users interact with API (jobs) directly in well-documented and
consistent way.
- Make controllers responsible only for mediating between views and API.
Behavior of these may remain inconsistent, since views they're talking
to are also messy to begin with. Such controllers might be removed
altogether in the future in favor of making views talk to API directly
through previously mentioned ApiController.
- Organize all sorts of privilege checking and possibly other stuff into
methods within jobs.
- Actually distinguish POST from GET requests.
- Leave POST-only controller methods as Actions, but rename GET-only
methods to Views. Example: editAction for editing comments, but
listView for showing comment list. The choice of these suffixes might
be subject to changes in future.
- Get rid of ?json and $context->transport. They now look like disease
to me.
This commit introduces job system and converts CommentController to use
the new API.
Instead of showing comments chronologically, group them into posts, then sort
the posts by last comment date. Reason: improved comment context delivery
makes discussion bumping possible (no matter how old it is) and discussion is
what comments are about.
Comment count is limited to 5 per post.
If user is in mass tag mode and changes target tag but doesn't change the
query, he now remains at the same page. (Concerns only users who have disabled
endless scrolling.)
Reason: until now, PostSearchService was using magic to get around the biggest
limitation of SqlQuery.php: it didn't support arbitrary order of operations.
You couldn't join with something and tell then to select something from it.
Additionally, forging UPDATE queries was a joke. The new Sql* classes replace
SqlQuery completely and address these issues. Using Sql* classes might be
tedious and ugly at times, but it is necessary step to improve model layer
maintainability.
It is by no menas complete implementation of SQL grammar, but for current needs
it's enough, and, what's most important, it is easily extensible.
Additional changes:
* Added sorting style aliases
- fav_count
- tag_count
- comment_count
* Sorting by multiple tokens in post search is now possible
* Searching for disliked posts with "special:disliked" always yields results
(even if user has disabled showing disliked posts by default)
* More maintainable next/prev post support
Following privileges for post actions can now understand different settings for
everyone and for uploader:
* Scoring posts
* Featuring posts
* Flagging posts
* Favoriting posts
Additionally, privilege for flagging users can now understand different
settings for everyone and for the user that is currently logged in.
In other words: with this update admin can configure privileges so that scoring
own posts or flagging oneself will be prohibited, while scoring other people's
posts or flagging others will be okay.
