Page MenuHomePhabricator

VE spontaneously adds numeric names to refs in sections not touched by a user edit
Open, Needs TriagePublicBUG REPORT

Description

Steps to replicate the issue (include links if applicable):

  • Start with a copy of Transgender history rev. 1085670210 of 20:16, 1 May 2022
  • Notice the scope of colon-numeric-named refs (":1", ":2", ":3", ":4")
  • Redo edit 1085883126 of 23:38, 2 May 2022 in VE, adding only the text shown in new section "Transgender People and the Medieval Church" and save
  • Note the nine new numeric refs added throughout the article (":5" .. ":13")

What happens?

  • Undesirable numeric named refs added to portions of the article not touched by the editor, changing content not intended by the editor, to the detriment of other editors who may wish to work on those sections.

What should happen?

  • Do not add numeric-named references not intended by the user.

Software version (skip for WMF-hosted wikis like Wikipedia):

  • en-wiki, vector-2010 (but irrelevant)

Other information (browser name/version, screenshots, etc.):

  • all/irrelevant

Event Timeline

Note the nine new numeric refs added throughout the article (":5" .. ":13")

Do you see this happening in the "review your changes" diff before you submit, or does it only crop up after saving?

Will have to run a test and see (prob. not today). In the meantime, perhaps the last three edits by Bowmanbk (1085915390, 1085883126, 1085878615) may provide a clue.

Given that these edits were in May, can you (@Mathglot) reproduce this with the latest version of Parsoid/VE before we dig into it more? (Also checking the "review your changes" diff as @DLynch requested.)

Looking at the user's contributions, it's clear that they copy-pasted the added content from their sandbox (https://en.wikipedia.org/wiki/Special:Contributions/Bowmanbk). I don't know how that caused the problem, but it's got to be related to that. It reminds me of T312050: VisualEditor changed reference names when reusing/copying references.

I tried to reproduce the problem, but cannot. You can see the attempt in User:Mathglot/sandbox/PhabT323419
(diff).

I don't think this necessarily means the problem is not reproducible, but that my technique did not reproduce it. The crux of the matter, imho, is in executing bullet 3 "Redo edit 1085883126". What I did, was to use the source editor to retrieve the wikicode of the new section Transgender People and the Medieval Church out of rev. 1085883126, and then edit User:Mathglot/sandbox/PhabT323419 using VE and pasting it into proper position above the "Balkans" section.

But the original editor pasted from User:Bowmanbk/Transgender history, which has three ":0" and two ":1" refs in the subpage. At a first glance, it looks like VE transforms numeric references contained in pasted text into different numeric reference numbers, if required; e.g., following the fragment "deviation from it" in the sandbox we have the code "<ref name=":0">Spencer-Hall, Alicia...</ref>", and in rev 1085883126 of TgH (after the paste) we have "<ref name=":1" />", and after "blessed by God" in the sandbox we have "<ref>Bychowski, Gabrielle...</ref>" and in rev 1085883126 of TgH we have "<ref name=":4" />". So a "paste" in VE isn't just a paste, it's a rejiggering of numeric named refs in some manner, even just considering that one subsection.

As to the original edit 1085883126 also rejiggering the numeric ref naes in other sections, I haven't been able to reproduce that yet, but I'm aware that my procedure was probably not exactly the same as what the user did. I should add that I'm not knowledgeable about VE, and there may be plenty of conventions or tricks of usage that VE editors use all the time, that I'm not aware of. So, I may not be the best person to try and reproduce this.

Probably I (or someone) should try a second test, pasting out of the user's sandbox, instead of out of the wikicode of the subsequent version, as I did. Feel free to reuse my sandbox subpage as a test bed, revert the last edit and try again, or whatever you want.

JMcLeod_WMF subscribed.

Content Transform Team does not think this is Parsoid-related, but should there be any evidence indicating it is, we will revisit it.