RFC: Disable caching for media filter

Here is the patch with the cache disabled.

Patch at #1 by aarom seems to work.
Here is the same patch for the 7.x-2.0-unstable version.

This patch should go into its own issue against the 2.0-unstable7 version. Here is the issue for the correct patch: https://drupal.org/node/2017761

#1: 1984424-disable-filter-caching-1.patch queued for re-testing.

Can I take that as RTBC?

A quick note to anyone trying this patch out:

You need to turn off the "Convert media tags to markup" filter in each of your textformats under admin/config/content/formats. Then turn it on again and save.

I did not get the patch working for a long time with or without cache enabled until I did the following above. This could save time for anyone out there trying to solve this issue.

That sounds like we need to do that in an update, then.

I strongly disagree with turning off caching of the media filter. I have run into several scenarios in the past 6 months where a 'cache' => FALSE setting in hook_filter_info() has brought a production site to its knees, causing all sorts of unexpected slowness and out of memory errors. To quote the hook_filter_info documentation:

cache (default TRUE): Specifies whether the filtered text can be cached. Note that setting this to FALSE makes the entire text format not cacheable, which may have an impact on the site's overall performance.

Please see the latest patch here: #1266064: The media filter allows infinite recursion via media_token_to_markup()

Here we go after some hair-pulling.

Patch looks good and it does what we wanted it to do.

I can, however, also agree with @kaidjohnson. Filtering text on every page request just because of this does seem a bit too much. Would it be possible to catch events on files (update, ...) and clear appropriate item from cache_field when needed? We should be able to do that since we have usage information for files that are embedded into WYSIWYG.

Thoughts?

That would be nice, except that we do not have any context information from the WYSIWYG. For example, if we have embedded a file into the body of node 21, and then later make a change to that file, we have no way currently to clear the cache for that body's text field. Although I suppose that we could clear all the caches every time we made a change to a file, but that seems a bit of an overkill to me. I guess it's six of one, half a dozen of the other.

#11: 1984424-disable-filter-caching-10.patch queued for re-testing.

Cache ID looks like this:

$cache_id = $format->format . ':' . $langcode . ':' . hash('sha256', $text);

Could we do this?
- hook on file_update
- check file usage for this file to get list of entities that use this file
- fetch list of textarea fields on those entites
- use field's content to calculate cache ID
- clear it

It is a bit funky, but probably still less overhead than making entire text format uncachable. We could, alternatively, flush all items that belong to formats that use media filter. Still less overhead than running filters over and over again, I assume.

That sounds reasonable.

File field does it's own caching of check_markup(), so the markup is cached in cache_field instead. That would cut out some of the work in #15.

Not file field, text module.

An updated version of #21 to use file_entity_invalidate_field_caches() and to remove the existing calls to media_filter_invalidate_caches() which essentially does the same thing as we're trying to accomplish but with the {media_filter_usage} table.

The patch also includes an update function to remove the table.

An updated version of #22 that removes the cache clearing from Media since it now exists in File entity.

Forgot to merge in the latest changes.

Committed #24 to Media 7.x-2.x.