Parsing/Replacing Tidy/FAQ

What is Tidy? edit

Tidy is a library formerly used by MediaWiki to fix some HTML errors found in wiki pages. Badly formed markup is common on wiki pages when editors use HTML tags in templates and on the page itself. (Example: unclosed HTML tags, such as a ‎<small> without a ‎</small>, are common). In some cases, MediaWiki can generate erroneous HTML by itself. Tidy fixes these markup errors, but also does other "cleanup" on its own that is not required for correctness. For example, it removes empty elements and adds whitespace between HTML tags, which can sometimes change rendering. Since Tidy is based on HTML4 semantics and the web has moved to HTML5, it also makes some incorrect changes to HTML to 'fix' things that used to not work; for example, Tidy will unexpectedly move a bullet list out of a table caption even though that's allowed.

Why are you replacing it? And with what? edit

Tidy's technology is from the 1990s, when browsers weren't standardized. Tidy's behavior is loosely based on HTML4 semantics but matches no modern browser. After spending years without active maintenance, Tidy has now been revived as "tidy-html5" with very different behavior. The older Tidy is no longer being packaged. As noted earlier, Tidy does HTML cleanup unrelated to fixing errors. Together, all these issues have led to lots of bugs filed against it on Phabricator, and a replacement has been asked for since at least 2013. HTML5 is the standard today, and the parsing algorithm for HTML5 is clearly specified, which has led to compatible implementations across browsers and other libraries. This algorithm also clearly specifies how broken markup should be fixed. In this new technological landscape, Tidy should really be replaced with an HTML5 parser that fixes up the broken markup and generates valid, well-formed HTML markup in the standard way. However, Wikimedia wikis have a huge corpus of pages whose markup relies on Tidy's fixups. Doing an immediate and straight-forward replacement of Tidy with a third-party HTML5 based tool is not feasible, since an HTML5-based tool would repair some markup differently and this can break how pages look. So, we are replacing Tidy with our own tool based on the HTML5 specification, but which also adds a few Tidy-compatibility workarounds to minimize the impact of replacing Tidy. After experimenting with 3 different solutions, we have settled on RemexHTML, a PHP-based HTML5 parser on top of which we have written Tidy-compatibility passes to replicate some Tidy behaviour which we need to provide for now. In the future, RemexHTML could also be used to enable new core MediaWiki features, such as more-robust section editing, balanced template support, and more efficient page updates after templates have been edited. For those who are wondering, note that using tidy-html5 would not preclude us from having to deal with fixing markup errors, since some of the required cleanup is due to changing from HTML4 to HTML5 semantics. There are other change-management reasons for preferring an in-house tool, including the ability to enable other features as mentioned above.

If you're interested in other technical details, please see phab:T89331 or Replacing Tidy.

Which tests have you performed so far? edit

To identify the impact of replacing Tidy with an HTML5-based tool, we have utilized a testing strategy (using a tool called "VisualDiff") that compares the pixel-by-pixel output image of MediaWiki with Tidy enabled, with the pixel-by-pixel output image of its replacement. Early on, we found that a common difference was minor vertical whitespace changes. In the belief that these would either not be noticeable or would be tolerable, we wrote a tool called "UprightDiff" which is able to identify vertical motion within an image and to discount such motion for the purposes of automated testing. This also let us assign a numeric score to differences and readily identify the most egregious differences. We exported a subset of about 64,000 articles (some from the recent changes stream, and rest selected randomly) from various Wikimedia wikis (40 wikis from Wikipedia, Wikisource, Wiktionary, and Wikivoyage), and rendered them with Tidy and with RemexHTML, then used "UprightDiff" to analyse the result. This takes a lot of cpu cycles, memory, and disk space, and it takes 2 days for one round of testing to complete. This limits the size of the testing corpus but we believe 64K is a sizeable sample to figure out the kind of fixes necessary.

To minimize the differences and reduce the impact of fixes that would be needed from editors, we added some additional Tidy-compatibility fixups. Since we found that self-closing tags were extremely common on wikimedia wikis, we added a compatibility fix to treat them as empty tags (i.e. ‎<b /> is treated as ‎<b>‎</b>). We added some other compatibility passes as well. After all these fixes were in place and we repeated our tests, we found that 93.4% of pages had no changes in rendering. And, 96.9% of pages had either no pixel diffs (93.4%) or insignificant vertical whitespace shifts only (3.5% = 96.9 - 93.4). The remaining 3.1% pages (100 - 96.9) showed pixel differences that had other reasons.

Based on these tests, we identified several classes of markup errors that will render differently between the two. For one class of markup errors (self-closing tags that aren't valid in HTML5), we added a maintenance category that editors have already been using to fix up templates and pages. But, the other classes of markup errors are not easy to detect automatically at this time and editors' assistance is necessary to identify and fix them up.

What will happen? When will these changes happen? edit

As noted earlier, we cannot yet do a drop-in replacement of Tidy with an HTML5-based tool. We have added a maintenance category for one class of markup errors, which will help editors fix them. To help editors identify and fix other markup errors, we also built a ParserMigration extension that helps them compare output in production and fix markup errors. Separately, we have also built the Linter extension to identify some of the fixes that are needed.

As of the end of March, 2017, we deployed the ParserMigration extensions to all wikis. As of June 20, 2017, Linter has been deployed to all large wikis. Via these extensions, we hope to enable editors to fix pages for actually replacing Tidy in 2017. Once enough fixes are made and we are satisfied that the impact on editors and readers will be minor and tolerable, we will go ahead and replace Tidy; however, we would also not like to drag this on indefinitely. So, it would be ideal if the high-priority issues identified by the Linter extension are prioritized by editors.

Separately, the Tidy-compatibility fixups (mentioned in the previous section) are meant to be in place till we replace Tidy. After that, we will start replacing these fixups gradually while relying on similar testing and tooling support.

What will editors need to do? edit

The Linter extension has been deployed to all wikis. As indicated on the help page, please fix wikitext patterns and templates identified in the high priority categories on the Special:LintErrors page of your wiki. Every item in that category has a help page with examples indicating what needs fixing. See simplified instructions below.

To assist editors in migrating wikitext to verify the fixes they make, we have deployed the ParserMigration extension. If you enable "ParserMigration" in your preferences, a link will be added to the toolbox of all articles, which can show the current (Tidy) and expected (RemexHTML) output side-by-side. You can preview article text changes in the same side-by-side view and see how your edits change / fix the rendering.

What is the impact on my wiki? edit

Please see the wikitext deprecation tool for this information. Note that in several cases the count refers to pages affected, but that includes template-generated issues. In other words, once a given template is fixed, all the pages featuring that template will disappear from the list. So it's highly likely that you won't have to fix thousands or millions pages, but a handful of templates. We are working to make this even more clear. You can also see some progress tracked via Parsing/Replacing Tidy/Linter/Stats, and the results of visual differences tests on a sample of ~73K articles from 60 wikis at .

Simplified instructions for fixing pages edit

Here are some simplified instructions for handling all the high-priority linter categories. Some of the linked help pages may contain instructions to use assisting tools such as WPCleaner.

Delete OR fix badly nested tables edit

{| <-- Table 1 starts here
| foo
{| <-- Table 2 starts here. You can remove this code.
|- <-- Row for Table 2. You may remove this too if you wish so.
| bar
|} <-- this closing tag is now useless and should be removed too

In this example, Tidy will delete Table 2 above. But, RemexHTML will not delete that table. This can change how pages look. To prevent this, editors should fix the wikitext and remove Table 2. While the following row-tag need not be removed, we recommend removing it. Since the closing table tag is no longer needed, it should be removed as well.

Alternatively, add an explicit | cell on the row started by the previous line before the start of Table 2 if you need nested tables. What is the correct fix depends on the page. But, in most cases deletion as above is going to be the right fix.

Work around a parser bug for paragraph wrapping edit

On most wikis, it looks like the biggest generator of these linter warnings are the nowrap or nowrap begin templates. The simplest fix that will handle the vast majority of these linter cases is to add a newline before the opening ‎<span> tag in the template source of these templates.

In all other cases, when wikitext has a span next to a div/td (and other such "block" tags) and has a white-space:nowrap CSS property, please add a newline after the div tag.

<div><span style='white-space:nowrap'>      <!-- <== ADD NEWLINE AFTER THE <div>. -->
foo bar
Note that this has the effect of enclosing the whole span in a paragraph element. Here the bug comes from newlines within the div element that cause a new paragraph to be generated for "foo".
The alternative, if you don't want a paragraph element automatically inserted (with its additional margins) by MediaWiki to surrounding the "span" inside the "div", is to not use any newline at all in the content of the "div" element, or to "hide" these newlines within HTML comments:
<div><span style='white-space:nowrap'>foo bar</span></div>
--><span style='white-space:nowrap'><!--
-->foo bar<!--

Fix invalid self-closing tags edit

Self-closing tags like ‎<div />, ‎<span />, ‎<b />, etc are not valid in HTML5. They need to be fixed according to what the editor intent might have been. In some cases, it is a typo where a ‎</b> is intended. In other cases, they need to be deleted. In some other cases, they need to be replaced with a ‎<nowiki />. Please see the detailed help page for this category.

Fix pages affected by a Tidy whitespace bug edit

Input wikitext Tidy output Remex output Proposed wikitext fix
<span class='nowrap'>a </span><span class='nowrap'>b</span>
<span class='nowrap'>a</span> <span class='nowrap'>b</span>
<span class='nowrap'>a </span><span class='nowrap'>b</span>
<span class='nowrap'>a</span> <span class='nowrap'>b</span>

Fix HTML5 vs Tidy misnested tag problems edit

Here is an example to illustrate the problem.

Input wikitext Tidy output Remex output Proposed wikitext fix



That was just one example to demonstrate the problem. There are other instances - ''foo<sup>''bar</sup> (an enwiki talk page) or <span>\n*x</span> (many itwiki pages via the use of citazione necessaria template) are other instances.

font tag with color attribute wrapping wikilinks edit

Here is an example to illustrate the problem.

Wikitext Tidy Remex Proposed Fix
<font color="green">[[Foo]]</font> <a href=".."><font color="green">Foo</font></a> <font color="green"><a href="..">Foo</a></font> [[Foo|<font color="green">Foo</font>]]
or better yet,
[[Foo|<span style="color:green;">Foo</span>]].

NOTE: Only a small set of font tags used on wikis are affected as explained above. All other uses don't need to be fixed. See the help page below for more details.

Multiple unclosed formatting tags edit

Wikitext Proposed Fix
<s>foo<s> bar
<s>foo</s> bar

Multi-line HTML table in lists edit

Wikitext Tidy RemexHTML Proposed Fix
* <table>

Unclosed quotes in heading edit

Wikitext Proposed Fix

Caveats edit

  • We don't yet know how to automatically determine all affected pages and templates, but we expect that the list of templates and suggestions will get more complete as we learn about problems that were not caught before. Community Liaisons will reach out to template experts, regulars at technical village pumps, etc. to make sure they become aware of the necessary changes and of resources that may help them.
  • We are thinking about other easy, sustainable ways in which we may support "gnomes" and all the Wikimedians willing to help with this effort.

Other FAQs edit

Q: What happens to ‎<pre /> and similar tags if they are used in self-closed form?

A: As noted in T134423, the only valid self-closed HTML tags are: area, base, br, col, embed, hr, img, input, keygen, link, meta, param, source, track, wbr. Non-HTML self-closing tags (like ‎<references /> and ‎<nowiki />) are not affected by this change. ‎<pre> is a special case because while it is an HTML tag, MediaWiki treats it like an extension tag and hence remains unaffected. All other self-closing HTML tags should be fixed (and are already being fixed by editors at this time).

Since this usage is found in a lot of pages in our testing, in order to prevent unexpected rendering effects (e.g. ‎<b /> being treated as ‎<b> and causing more text being bolded than intended), we added a fix to the parser to convert them to an empty tag (eg. ‎<b /> will be converted to ‎<b>‎</b>). But, we don't intend to retain this fix indefinitely. So, we would like for editors to continue fixing this deprecated usage.

Q: What were the results of tests on languages other than English, or on sister projects?

A: There is nothing in what Tidy, RemexHTML do that is specific to Wikipedia or English. This project is primarily about a change from HTML4 to HTML5 semantics and getting rid of some Tidy cleanups of HTML. These changes affect all projects and languages equally, except if some projects and languages tend to have more markup errors or use more self-closed tags than others.

Q: What other changes are editors likely to see after this replacement?

A: The effect of this replacement is primarily going to affect readers, as they may notice that the page doesn't look right (for example, excessively wide navboxes without line breaks or wrapping). However, if anything, this might lead to the rendering seen in VisualEditor to match the rendering seen outside it much more than before, since Parsoid's output has been HTML5-compliant since the beginning, and we are now moving the read output to HTML5. We do not expect any impact on VisualEditor edits, but we will promptly address any bugs reported with respect to dirty diffs. In addition, we do not plan to add any error messages or warnings displayed on pages if the markup errors are not fixed.

Q: How does the replacement relate to other projects you are working on?

A: By enabling the move to HTML5 semantics, this is one of the steps evolving markup in our corpus to keep up with web standard. We also expect to leverage this tool to support well-balanced template output. Separately, but relatedly, this will also make the output of the PHP parser (used for reads) and the output of Parsoid (used for edits in VE, Content Translation) more consistent since Parsoid already uses HTML5 semantics. One of our goals is to make the two outputs fully consistent with each other and use one parser for both reads and edits.

--The Parsing Team

Volunteers available to support this effort edit

Community Liaisons invite interested Wikimedians to please add their name in the sections below and support their community engagement efforts. Thank you. As with similar past initiatives, signing up is optional. Please see Parsing/Get_involved, and add it to your bookmarks, as future requests for assistance will go through that page!

  1. I am available to test with the ParserMigration tool.
    1. (add your signature here)
    2. ...
    3. ...
  2. I am available to fix templates.
    1. Jonesey95 (talk) 16:10, 14 November 2016 (UTC)[reply]
    2. Samuele2002 (talk)
    3. TheDragonFire (talk)
    4. Stryn (talk) 14:22, 17 July 2017 (UTC)[reply]
    5. Already did some in fawiki Ladsgroup (talk) 15:40, 18 September 2017 (UTC)[reply]
    6. My bot can fix multiple-unclosed-formatting-tags error. --星耀晨曦 (talk) 07:19, 6 April 2018 (UTC)[reply]
    7. ネイ (talk) 09:36, 7 April 2018 (UTC)[reply]
    8. Been at it a long time already at en.Wikipedia.  — SMcCandlish ¢ ≽ʌⱷ҅ʌ≼  06:56, 3 May 2019 (UTC)[reply]
  3. I am available to study and discuss fixes to templates.
    1. Jonesey95 (talk) 16:10, 14 November 2016 (UTC)[reply]
    2. Been at it a long time already at en.Wikipedia.  — SMcCandlish ¢ ≽ʌⱷ҅ʌ≼  06:56, 3 May 2019 (UTC)[reply]
  4. I am available to spread the word among my community.
    1. (See this page) --Sannita (talk) 18:02, 8 July 2017 (UTC)[reply]
    2. See this page. Stryn (talk) 14:22, 17 July 2017 (UTC)[reply]
    3. See this page. ネイ (talk) 09:36, 7 April 2018 (UTC)[reply]
    4. See meta:User:SMcCandlish/lint.css; w:Linter#User CSS tool: lint.css; w:Template:Mxt/User CSS for a monospaced coding font; etc.  — SMcCandlish ¢ ≽ʌⱷ҅ʌ≼  06:56, 3 May 2019 (UTC)[reply]