Move smart_quotes feature to post_transform

[tk0miya]

Problem

• Not to cache the result of smart_quotes, the processing should be moved to post_transform.

Procedure to reproduce the problem

$ make html text
or
$ make text html

(with #4360)

Error logs / results

• Some outputs are applied smart_quotes or not applied.

Expected results

• Applied smart_quotes correctly under smartquotes_disables.

Note:

• This might requires a big refactoring. So it would be implemented 1.7 or later.

[CendioOssman]

I'm not sure what the issue is here? The transform doesn't look like it is doing terribly complex things. So what is preventing it from just being changed to a post transform?

We got hit by this by doing both man pages and html output with a shared doctree cache. And suddenly the order of things became very important.

[tk0miya]

At present, some transforms are invoked after smart_quotes transform and are collecting contents from the "smart-quoted" doctree. For example, the standard domain collects section titles, titles of anchors, and so on. If we'll move the smart_quotes transform to post transforming phase, these collected contents will not be smart-quoted.

To change the order of processing, we need to create a new mechanism to do smart-quoting the collected contents.

[CendioOssman]

Ah, that does complicate things. Most cases would hopefully be handled if smart quotes is run sufficiently late in the process though? E.g. reference titles.

And I don't suppose anyone has considered moving those collections to somewhere later? I know I had some issues with the fact that adding certain nodes in the post transform stage didn't work as the standard domain didn't collect information from those.

[tk0miya]

We also have similar problems for i18n feature. And we've not been able to find the better way to do them on at the later step yet.