If I understand the semantic meaning of rel="canonical" correctly, what it does now is the expected behavior.

http://zh.wikipedia.org/wiki/%E6%B1%89%E8%AF%AD is not "the Simplified Chinese
page", but an automatically converted page based on requests (prefs for users and Accept-Language for anons). We want all these links to show up in Google search results instead of links specifying a particular variant.

However Google seems not respecting it and indexing links to pages in every variant, and we have to workaround it: https://zh.wikipedia.org/w/index.php?title=MediaWiki:Gadget-variant-link-fix.js

Currently Google seems to mainly index /zh/ links instead of /wiki/'s (which is unexpected). /zh-tw/ or /zh-cn/'s are not indexed as expected though.

RobLa: So should this still be high priority wrt Liangent's comment 1 here?

If still high priority:
Tim: Do you plan to work on this at some point?

(In reply to Andre Klapper from comment #3)

RobLa: So should this still be high priority wrt Liangent's comment 1 here?

If still high priority:
Tim: Do you plan to work on this at some point?

I guess Tim is just the default CC, but actually this issue seems not Wikimedia-specific.

Change 154240 had a related patch set uploaded by Tim Starling:
Don't send rel=canonical to variant-neutral page

https://gerrit.wikimedia.org/r/154240

Change 154240 merged by jenkins-bot:
Don't send rel=canonical to variant-neutral page

https://gerrit.wikimedia.org/r/154240

All patches mentioned in this report were merged or abandoned - is there more work left to do here (if yes: please reset the bug report status to NEW or ASSIGNED), or can you close this ticket as RESOLVED FIXED?

(In reply to Rob Lanphier from comment #0)

Language variants currently point to the same canonical URL. For example, on
this page:

http://zh.wikipedia.org/zh-tw/%E6%B1%89%E8%AF%AD

...there is a rel=”canonical” pointing to
http://zh.wikipedia.org/wiki/%E6%B1%89%E8%AF%AD

Now has:

<link rel="alternate" hreflang="zh" href="/zh/%E6%B1%89%E8%AF%AD" />
[...]
<link rel="alternate" hreflang="zh-TW" href="/zh-tw/%E6%B1%89%E8%AF%AD" />
<link rel="alternate" hreflang="x-default" href="/wiki/%E6%B1%89%E8%AF%AD" />
[...]
<link rel="canonical" href="http://zh.wikipedia.org/zh-tw/%E6%B1%89%E8%AF%AD" />

But I'm not sure this is properly fixed in general, because this is still an issue:

(In reply to fireattack from comment #2)

Currently Google seems to mainly index /zh/ links instead of /wiki/'s (which
is unexpected). /zh-tw/ or /zh-cn/'s are not indexed as expected though.

The two URLs for "zh" version don't agree on which is canonical:

/zh/ says

<link rel="alternate" hreflang="zh" href="/zh/%E6%B1%89%E8%AF%AD" />
[...]
<link rel="canonical" href="http://zh.wikipedia.org/zh/%E6%B1%89%E8%AF%AD" />

/wiki/ says

<link rel="alternate" hreflang="zh" href="/zh/%E6%B1%89%E8%AF%AD" />
[...]
<link rel="canonical" href="http://zh.wikipedia.org/wiki/%E6%B1%89%E8%AF%AD" />

Created attachment 16795
Google search in Italian for [[zh:汉语]]

If I search a Latin alphabet string of that article I manage to get 4 variants from Google after asking to show me duplicate pages as well. None of them is /wiki/

Searching '"漢語，又称中文、华语" site:wikipedia.org' yielded two results including zh.wap.wikipedia.org/zh-tw/汉语 but that's another bug.

Attached:

I don't know why the title of this page reads "Canonical URL should include language variant" (because it shouldn't). But anyway, I'm here to report the exact weired behavior mentioned above:

In T54429#541103, @Nemo_bis wrote:

Created attachment 16795
Google search in Italian for [[zh:汉语]]

If I search a Latin alphabet string of that article I manage to get 4 variants from Google after asking to show me duplicate pages as well. None of them is /wiki/

Searching '"漢語，又称中文、华语" site:wikipedia.org' yielded two results including zh.wap.wikipedia.org/zh-tw/汉语 but that's another bug.

Attached:

If you use google to search "汉语 维基百科", the first result would be https://zh.wikipedia.org/zh/汉语 and all the other results below are using /zh/.

This is NOT optimal because it will show page in original variant instead of user's preference (in my case, zh-cn).

However if you search with "汉语 维基百科 site:wikipedia.org", the first result will become https://zh.wikipedia.org/wiki/汉语 as well as other links. This is optimal because /wiki/ links would automatically jump to language variant that user wants.

I have no idea what causes this strange situation (it's even maybe Google's fault), but it needs to be fixed. It's quite annoying that users need to manually change language variant from Google result.

So I think this bug should be "Canonical URL should be /wiki/ links, but somehow Google doesn't honor it".

Change 609513 had a related patch set uploaded (by VulpesVulpes825; owner: VulpesVulpes825):
[mediawiki/core@master] Write language varaint link as child element rather than individual entry in sitemap

https://gerrit.wikimedia.org/r/609513