[RFC] Integrate s9e\TextFormatter

[JoshyPHP]

I went with the original plan B: query the database, glob all the styles/*/template/bbcode.html files and recompose BBCode templates based on the fragments. The commit is here. The path to styles/ is configurable. This is what the renderer looks like for prosilver+subsilver.

[EXreaction]

Styles have inheritance, so you'll need to check the inheritance tree as well to find the appropriate file.

[JoshyPHP]

I assume that styles inherit the whole bbcode.html, not just parts of it right? Do they also inherit the same bbcode_bitfield or can it be different from its parent?

Update: pushed to GitHub. I recurse into each style's parents until I find one that has a bbcode.html file. I don't touch the style's bbcode_bitfield, so if it's different from its parent they are renderered differently.

[EXreaction]

Yes, although I do not know what the bitfield is used for in the styles.

[JoshyPHP]

It determines which default templates are replaced. For instance, the default template for [b] uses <span style="font-weight: bold">. prosilver's template uses <strong>, but since the bit for [b] is not set, it uses the default template instead.

[Pony99CA]

When all BBCodes become "custom" BBCodes, maybe there should be a new Allow style override setting on the Custom BBCodes page instead of this bitfield.

Steve

[JoshyPHP]

Today's update: I've added support for the attachment BBCode. Once I made sense of how the current code works, I managed to adapt to it with minimum work.

For reference, here's how the legacy routines work:

1. The user posts [аttachment=0]icon_e_biggrin.gif[/attachment]
2. It is stored as [аttachment=0:16rvhx28]<!-- ia0 -->icon_e_biggrin.gif<!-- ia0 -->[/attachment:16rvhx28]
3. bbcode::bbcode_second_pass() renders it as <div class="inline-attachment">…<!-- ia0 -->icon_e_biggrin.gif<!-- ia0 -->…</div>
4. The result is post-processed by parse_attachments() to display the attached file as per the attachment.html template

And with the new routines, it's basically the same thing except step 3 is handled by the text_formatter.renderer service and in step 2 it is stored as:

Code: Select all

<rt><ATTACHMENT filename="icon_e_biggrin.gif" index="0"><st>[attachment=0]</st>icon_e_biggrin.gif<et>[/attachment]</et></ATTACHMENT></rt>

[JoshyPHP]

I've added some basic tests today. One of them (misnested tags) made me realize something (that phpBB uses span elements.) But first, a word about tag rules in s9e\TextFormatter.

s9e\TextFormatter supports a relatively wide array of rules governing tags, such as BBCode tags. Those rules allow you to control where tags can be used (e.g. you can't have a link inside of a link), what to do with their whitespace and/or newlines outside of inside of them or how to handle misnesting and optional end tags (e.g. [/*]). Fortunately, most of the time you don't have to configure any of that because a set of rules is automatically generated by studying the templates' content. This is done using the HTML5 specs as a guideline, and more specifically HTML5's content models, optional tags, adoption agency algorithm, and suggested default rendering. That's how it magically knows that links shouldn't be nested, that [/*] tags are optionals and that it shouldn't convert new lines to <br> inside of a <pre> tag. All spiffy stuff that I will be eternally proud of.

There are, however, some situations where the automatic rules may not cover all of the user expectations. For instance, if your templates use the b and i (or strong and em) elements for your [b] and [i] BBCodes, the following text:

Code: Select all

[b]..[i]..[/b]..[/i]

...will be interpreted as:

Code: Select all

[b]..[i]..[/i][/b][i]..[/i]

...which is exactly what the user wanted: everything between [b] tags is bold, everything between [i] tags is italic. That's because BBCodes that use the "b", "i", "em" or "strong" (and a few other) HTML elements, are automatically set to be reopened if they're closed unexpectedly. For instance, when their parent gets closed as in the example above. phpBB's prosilver uses a span element, which is not on the list of elements that get reopened automatically. That's why I explicitly added a rule to [b], [i] and [u] so that they get reopened automatically, because I think that's what people would expect.

There are other situations where I'm not sure what people expect. Take this bug for instance: PHPBB3-9386 - and BBCodes also affect quotes. A "quote" is not valid inside of a "b", so currently it results in bold, non-quoted text. I could set it so that quotes break the bold text, but I'd need feedback on that. Should "quote" close a "b"? If so, what other tags should it close? "i", "u", "color", I guess. "url" I'm not so sure.

[EXreaction]

I think we'd prefer less autocorrection/magic rather than try to make things overly complex.

[JoshyPHP]

I'm not sure what you mean there, or what your concern is. What do you think would become complex?

Taking PHPBB3-9386 as an example, the difference between the quote breaking bold/italic or the quote being disabled is one option. Either I pass 'closeParent' => array('b', 'i', 'u') as an option when creating the quote tag, or I do nothing and the default is for "quote" to be disallowed inside of "b". Or I can disable all rules altogether, in which case you get the same bold quote as in the bug report.

Having some sensible defaults in term of autocorrection saves the end user a lot of hassle. (e.g. nobody wants to close every list item with [/*]) What defaults would you want to have?