Context:

With the https://gerrit.wikimedia.org/r/1202126 we made a change that potentially affects a few use cases see also T408535: Look into the failing tests when rebuilding keyIndexes. While smoke testing this on the beta cluster I could see the following regressions in some edge cases.

Regressions:

• Start with the following Wikitext

<ref>Autoname Book</ref>

<ref name="ldr" />

<ref details="page 1" name="mainplus">Main+details</ref>

<references>
<ref name="ldr">LDR Book</ref>
</references>

• Open the page for editing in VE in one tab
• Open the page for editing in VE in another tab

Inline ref:

• Select and copy the first ref [1] from the 1st tab
• Pasted it twice next to the [1] in the 2nd tab

What happens:

• As soon as we paste the 2nd time one of the added footnotes shows as [undefined]

What should happen:

• Both footnotes should show a number, different from the original

LRD ref ( no regression ):

• Start over
• Select and copy the secound ref [2] from the 1st tab
• Pasted it twice next to the [2] in the 2nd tab
• This behaves as intended

Sub-ref:

• Start over
• Select and copy the third ref [3] from the 1st tab
• Pasted it twice next to the [3] in the 2nd tab

What happens:

• As soon as we paste the 2nd time one of the added footnotes shows as [undefined]

What should happen:

• Both footnotes should show a number, different from the original

Notes:

• Generally it seems that named references do not have that issue. It seems to affect reference nodes with auto/ listKey
• From some initial pocking around it seems that ve.dm.MWReferenceNode.static.remapInternalListKeys does not work here in all cases.