Le 23/07/2015 08:15, Ricordisamoa a écrit : You might want to explain what you are trying to do and which wall you have hit when attempting to use Parsoid :-)

HTML5+RDFa is a machine-readable format. But I think what you are asking for is either better documentation of the template-related stuff (did you read through the slides in https://phabricator.wikimedia.org/T105175 ?) or HTML template parameter support (https://phabricator.wikimedia.org/T52587) which is in the codebase but not enabled by default in production. --scott ​ _______________________________________________ Wikitech-l mailing list Wikitech-l(a)lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l

Well, it's really just a different way of thinking about things. Instead of: ``` ``` you would write: ``` js> Parsoid = require('parsoid'); js> text = "I has a template! {{foo|bar|baz|eggs=spam}} See it?"; js> Parsoid.parse(text, { document: true }).then(function(res) { templates = res.out.querySelectorAll('[typeof~="mw:Transclusion"]'); console.log(templates); }).done(); ``` That said, it wouldn't be hard to clone the API of http://mwparserfromhell.readthedocs.org/en/latest/api/mwparserfromhell.html

and that would probably be a great addition to the parsoid package API. HTML is just a tree structured data representation. Think of it as XML if it makes you happier. It just happens to come with well-defined semantics and lots of manipulation libraries. I don't know about edits tagged as "VisualEditor". That seems like that should only be done by VE.

I take it you would like an easy work flow to fetch a page, make edits, and then write the new revision back?

mwparserfromhell doesn't actually seem to have that functionality

, but it would also be nice to facilitate that use case if we can. --scott ​ _______________________________________________ Wikitech-l mailing list Wikitech-l(a)lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l

Il 23/07/2015 15:28, Antoine Musso ha scritto: For example, adding a template transclusion as new parameter in another template. XHTML5+RDFa is the wall :-( Can't Parsoid's deserialization be caught at some point to get a higher-level structure like mwparserfromhell <https://github.com/earwig/mwparserfromhell>'s?

Il 24/07/2015 06:35, C. Scott Ananian ha scritto: should only be done by VE. All edits made via visualeditoredit < https://www.mediawiki.org/w/api.php?action=help&modules=visualeditoredit are tagged.

On 24 July 2015 at 07:34, Ricordisamoa &lt;ricordisamoa(a)openmailbox.org&gt; wrote: RESTBase could help you there. With one API call, you can get the (stored) latest HTML revision of a page in Parsoid format~[1], but without the need to wait for Parsoid to parse it (if the latest revision is in RESTBase's storage).

There is also section API support (you can get individual HTML fragments of a page by ID, and send only those back for transformation into wikitext~[2]). There is also support for page editing (aka saving), but these endpoints have not yet been enabled for WMF wikis in production due to security concerns.

[1] https://en.wikipedia.org/api/rest_v1/?doc#!/Page_content/page_html__title__… [2] https://en.wikipedia.org/api/rest_v1/?doc#!/Transforms/transform_sections_t… Cheers, Marko

On 23 July 2015 at 22:34, Ricordisamoa &lt;ricordisamoa(a)openmailbox.org&gt; wrote: ​Yes. That's because that is the *private* API for VisualEditor. It absolutely should not ever be used by anyone else.​ It's not like any of the 'real' APIs in MediaWiki – it is designed for exactly one use case (VisualEditor), makes huge assumptions about the world and what is needed (like tagging edits), and we make breaking changes all the time. Unfortunately, the request to badge internal APIs got turned into flagging it and similar APIs in MediaWiki as "This module is internal or unstable.", which isn't strong enough on just how bad an idea it is to use it. I would extremely strongly suggest that you do not use it, ever.

As Marko, Subbu and Scott point out, we have actual public APIs for this kind of stuff, in the forms of RESTbase and Parsoid, and that's what you should use. Yours,

On 07/23/2015 01:07 PM, Ricordisamoa wrote: Parsoid and mwparserfromhell have different design goals and hence do things differently. Parsoid is meant to support HTML editing and hence provides semantic information as annotations over the HTML document. It effectively maintains a bidirectional/reversible mapping between segments of wikitext and DOM trees. You can manipulate the DOM trees and get back wikitext that represents the edited tree. As for useless information and duplicate information -- I think if you looked at the Parsoid DOM spec [1], you will know what to look for and what to manipulate. The information on the DOM is meant to (a) render accurately (b) support the various bots / clients / gadgets that look for specific kinds of information, and (b) be editable easily. If that spec has holes or needs updates or fixing, we are happy to do that. Do let us know. mwparserfromhell is an entirely wikitext-centric library as far as I can tell. It is meant to manipulate wikitext directly. It is a neat library which provides a lot of utilities and makes it easy to do wikitext transformations. It doesn't know about or care about HTML because it doesn't need to. It also seems to effectively gives you some kind of wikitext-centric AST. These are all impressions based on a quick scan of its docs -- so pardon any misunderstandings. Parsoid does not provide you a wikitext AST directly since it doesn't construct one. All wikitext information shows up indirectly as DOM annotations (either attributes or JSON information in attributes). As Scott showed, you can still do document ("wikitext") manipulations using DOM libraries, CSS-style queries, or directly by walking the DOM. There are lots of ways you can edit mediawiki pages without knowing about wikitext and using the vast array of HTML libraries. That happens to be our tagline: "we deal with wikitext so you don't have to". But, you are right. It can indeed seem cumbersome if you want to directly manipulate wikitext without the DOM getting in between or having to deal with DOM libraries. But that is not the use case we target. There are a vastly greater number of libraries in all kinds of languages (and developers) that know about HTML and can render, handle, and manipulate HTML easily than know how to (or want to) manipulate wikitext programmatically. Kind of the difference between the wikitext editor and the visual editor. They each have their constituencies and roles. All that said, as Scott noted, it is possible to develop a mwparserfromhell like layer on top of the Parsoid DOM annotations if you want a wikitext-centric view (as opposed to a DOM-centric view that most editing clients seem to want). But, since that is not a use case that we target, that hasn't been on our radar. If someone does want to take that on, and thinks it would be useful, we are happy to provide assistance. It should not be too difficult. Does that help summarize this issue and clarify the differences and approaches of these two tools? I am "on vacation" :-) so responses will be delayed. Subbu. [1] http://www.mediawiki.org/wiki/Parsoid/MediaWiki_DOM_spec

Hi Subbu, thank you for this thoughtful insight.

HTML is not a barrier by itself. The problem seems to be Parsoid being built primarily with VisualEditor in mind.

It is not clear to me how can a single DOM serving both view and edit modes avoid redundancy.

I see huge demand for alternative wikignome-style editors. The more Parsoid's DOM is predictable, concise and documented, the more users you get.

I hope we can meet in the middle :-)

On 07/31/2015 12:55 PM, Ricordisamoa wrote: And thank you for starting this thread. :-) While we want the DOM to be VE-friendly, we definitely don't want the DOM to be VE-centric and that has been the intention from the very beginning. Flow, CX also use the Parsoid DOM for their functionality. There are other users too [1].

We definitely want Parsoid's output to be useful and usable more broadly as the canonical output representation of wikitext and are open to fixing whatever prevents that. As Scott noted in the other email on the thread, inspired (and maybe challenged by :-) ) by mwparserfromhell's utilities, he has already whipped out a layer that provides an easier interface for manipulating the DOM. You are right that there are some redundancies in information representation (because of having to serve multiple needs), but as far as I know, it is mostly around image attributes. If there is anything else specific (beyond image attributes) that is bothering you, can you flag that?

I think Parsoid's DOM is predictable :-) but, can you say more about what prompted you to say that?

As for documentation, we document the DOM we generate and its semantics here [2].

As for size, I just looked at the Barack Obama page and here are some size numbers.

1540407 /tmp/Barack_Obama.parsoid.html 1197318 /tmp/Barack_Obama.parsoid.no-data-mw.html 1045161 /tmp/Barack_Obama.php-parser.output.footer-stripped.html Right now, because we inline template and other editable information (as inline JSON attributes of the DOM), it is a bit bulky. However, we have always had plans to move the data-mw attribute into its own bucket which we might at some point in which case the size will be closer to the current PHP parser output. If we moved page properties and other metadata out, it will shrink it a little bit more. For views that don't need to support editing or any other manipulation or analyses, we can more aggressively strip more from the HTML without affecting the rendering

and get close to or even shrink the size below the PHP parser output size (there might be use cases where that might be appropriate thing to do). I could get this down to under 1M by stripping rel attributes, element ids, and about ids for identifying template output. But, for editing (not just in VE) use cases, because of additional markup in place on the page (element ids, other markup for transclusions, extensions, links, etc.), the output will probably be somewhat larger than the corresponding PHP parser output. If we can keep it under 1.1x of php parser output size, I think we are good. Please file bugs and continue to report things that get in the way of using Parsoid. Subbu. [1] https://www.mediawiki.org/wiki/Parsoid/Users [2] http://www.mediawiki.org/wiki/Parsoid/MediaWiki_DOM_spec _______________________________________________ Wikitech-l mailing list Wikitech-l(a)lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l

You are right that there are some redundancies in information https://www.mediawiki.org/wiki/Parsoid/MediaWiki_DOM_spec#Transclusion_cont… All template parameters are in data-mw but not parsed. Parameters ending up in the 'final' wikitext are parsed separately.

I see huge demand for alternative wikignome-style editors. The more For example, to find images I have to search elements where typeof is one of mw:Image, mw:Image/Thumb, mw:Image/Frame, mw:Image/Frameless, then see if it's a figure or a span, and expect either a <figcaption> or data-mw accordingly. Add that the img tag's parent can be <a> or <span>... Instead, this is what I'd expect a proper structure to look like:

Il 03/08/2015 22:08, C. Scott Ananian ha scritto: I'm baiting you into exposing a mwparserfromhell-like AST from RESTBase. Then I can deal with a Python client, a PHP one, etc. :-) I assure you there is nothing JavaScript-specific about this; there are _______________________________________________ Wikitech-l mailing list Wikitech-l(a)lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l

Good news: https://doc.wikimedia.org/Parsoid/master/#!/guide/jsapi now documents the new friendlier API for Parsoid. --scott On Tue, Aug 11, 2015 at 3:03 PM, Ricordisamoa &lt;ricordisamoa(a)openmailbox.org&gt; wrote:

Stephen Niedzielski: "it seems like, as soon as you get the HTML the first thing you want to do, perhaps a little bit ironically because it's called Parsoid, it's parse the output a little bit more" https://www.youtube.com/watch?v=3WJID_WC7BQ&t=35m14s Il 23/07/2015 22:02, C. Scott Ananian ha scritto: _______________________________________________ Wikitech-l mailing list Wikitech-l(a)lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l