Previously(?) newly-created links (without data-parsoid) of either type –
| <p><a rel="mw:ExtLink" href="http://en.wikipedia.beta.wmflabs.org/wiki/European_Robin">European Robin</a></p> |
… and …
| <p><a rel="mw:WikiLink" href="http://en.wikipedia.beta.wmflabs.org/wiki/European_Robin">European Robin</a></p> |
… – would both serialise to
| [[European Robin]] |
but now the latter is serialising to
| [http://en.wikipedia.beta.wmflabs.org/wiki/European_Robin European Robin] |
[subbu@earth lib] cat /tmp/html
<a rel="mw:WikiLink" href="./Foo" title="Foo" data-parsoid='{"stx":"piped","a":{"href":"./Foo"},"sa":{"href":"Foo"},"dsr":[0,11,6,2]}'>Bar</a>
<a rel="mw:WikiLink" href="./Foo" title="Foo">Bar</a>
<a rel="mw:WikiLink" href="http://en.wikipedia.org/wiki/Foo" title="Foo">Bar</a>
<a rel="mw:ExtLink" href="http://en.wikipedia.org/wiki/Foo" title="Foo">Bar</a>
[subbu@earth lib] node parse --html2wt < /tmp/html
[[Foo|Bar]]
[[Foo|Bar]]
[[http://en.wikipedia.org/wiki/Foo|Bar]]
[[Foo|Bar]]@subbu: shoud mw:ExtLink be treated differently from mw:WikiLink, or should they be treated as synonyms?
In your example, it seems like the last two cases are exactly backwards. Maybe there's just a ! missing somewhere...
I am not sure why an editor would add a mw:ExtLink annotation on a wikilink, but at this time, I think we treat mw:ExtLink / mw:WikiLink as hints for the most part.
I am not sure if there is a "!" missing .. because all of them are being serialized to wikilink syntax. With a misplaced ! I would have expected an extlink syntax.
Change 218405 had a related patch set uploaded (by Cscott):
T94723: Fix serialization of mw:WikiLink which use absolute URLs
See T102556. Our current implementation (with the gerrit patch above) is fragile, since it does not let users actually author an ExtLink to a WMF property if they wanted to. This shows up as selser failures:
[http://en.wikipedia.org/wiki/Foo Bar] round-trips correctly only when unmodified (ie, via selser).
As soon as it is edited (say Bar becomes Baz), it turns into [[Foo|Baz]] (not [http://en.wikipedia.org/wiki/Foo Baz]) which is a selser failure.
The only way around this is to flag the URL as "really an ExtLink, don't make it a Wikilink" somehow. T102556 provides a means to do this, but I'm not sure it's worth it.
Change 218405 merged by jenkins-bot:
T94723: Fix serialization of mw:WikiLink which use absolute URLs
Update, using production-deployed parsoid-lb interface:
<p><a rel="mw:ExtLink" href="https://en.wikipedia.org/wiki/European_Robin">European Robin</a></p>
still maps to
[[European Robin]]
… but …
<p><a rel="mw:WikiLink" href="https://en.wikipedia.org/wiki/European_Robin">European Robin</a></p>
now maps to [[https://en.wikipedia.org/wiki/European Robin|European Robin]] which is rather broken. :-(
@Jdforrester-WMF That patch hasn't been deployed yet.
But this looks ok on master,
λ (master) echo '<p><a rel="mw:WikiLink" href="https://en.wikipedia.org/wiki/European_Robin">European Robin</a></p>' | node parse --html2wt [[European Robin]]
Yes, the broken [[https://en.wikipedia.org/wiki/European Robin|European Robin]] serialization is one of the things this patch specifically fixes. (I believe there's even a test case for it in the patch.)