Bibliography CSL "substitute" element does not always work as expected

[Omikhleia]

When a <csl:substitute> element is processed, "substituted variables are suppressed in the rest of the output to prevent duplication"

Well, our logic here is somewhat broken :(

Sometimes they are duplicated, sometimes there's too much suppression. It highly depends on the CSL style (i.e. how it has nesting, groups, choose conditions etc.), so an MWE is hard to extract.

I probably introduced a regression via commit e805a2c (so since 0.15.10) which affects some of my workflows (= duplicated items, due to wrapping macros in a pseudo-groups) with my own styles.

But the logic was likely already broken from the origin... Currently, we might even suppress a variable at sorting, and we then don't have it any longer at rendering... Oops. It's not very clear how to behave here. But even cancelling suppression at sorting, some problem remain afterwards.
I also experimented a quick hack with my main (custom MLA-based) style, but it's probably not a generic fix, and I'm afraid we'd have to track substitutions better than we currently do, the initial stuff was naive there...

So I am just reporting the issue for now, at least, with no idea yet how to fix it properly.
One of the problems is that we don't have tests yet for this CSL light implementation to check both for correctness and regression, but crafting them would be hard (and the CSL tests in their repository are fairly messy too to start with....).

[Omikhleia]

For the record here, some of the "problems" are repaired in #2294 but the changes might not be general enough to address all cases. I still need to work on a MWE for some other styles that I don't use as much as Chicago¹ or MLA, with entries that would trigger the issue and help understanding where the implementation errs.

Footnotes

1. As we have it ("vendored") in our package. The style on the CSL repository was recently upgraded (to a newer edition of the Chicago Manual of Styles) and seems problematic (at a quick glance). It uses a lot more nested constructs in macros and name substitutes than the previous version, so it's not the easiest thing to debug ^^ ↩