After comparing Stylesheets in release 7.56.0 (which work as expected) and 7.57.0 (which does not), I have found that the difference causing the issue is in /source/p5subset.xml.

So there is nothing to blame in the Stylesheets (at least in terms of changes made for the last stable release). What has changed between release 7.56.0 and 7.57.0 that is causing the issue is p5subset.

p5subset in 7.56.0 has

<elementSpec module="tagdocs" ident="tag">
  ...
  <attDef ident="type" mode="change" usage="opt">
     ...
     <valList type="closed" mode="add"> 

While p5subset in 7.57.0 does not have mode="add" on the valList, which is causing it to fall through the cracks

So we either make sure that implicit mode="add"s are processed correctly, or we restore explicit mode="add"s in the places where we removed them.

Excellent, thank you @raffazizzi!

As for the two proposed solutions, seems to me explicitly adding mode="add" in our declarations in the Guidelines source is a relatively easy fix we can implement almost right away. But that won’t fix the same problem occurring in other people’s ODDs, so I think we need to fix processing of implicit mode="add" whether we implement the easy fix for P5 or not. (That is, changing our declarations does not relieve us of the obligation to fix the processing, but it does mean we could fix it at a normal pace, rather than as a hot-mess, high-priority, emergency.)

The good news is that in this case (as far as I can tell) there is no problematic conflict between the declaration of @mode and the prose description of it. (As there sort of is for some other cases.)

In this case, att.combinable defines the default value of @mode to be "add"; and chapter TD explicitly says that mode="change" means “process identifiable children according to their modes; process unidentifiable children in replace mode; retain existing children where no replacement or change is provided”.

Thus if @mode is missing on a child of something that has @mode="change" (and for which there is another specification), the default of "add" should just be honored.

I have taken a look (using [1]), and it seems as though there are 44 cases of missing @mode="add" on <valList> inside <attDef mode="change"> in the current p5subset, all on re-definitions of @type.

[1]

for path in $(xsel -t -m "//t:attDef[@mode='change'][t:valList[@mode='add']]" -v 'concat("//t:elementSpec[@ident=%", ../../@ident,"%]/t:attList/t:attDef[@ident=%", @ident, "%]/t:valList")' -n Stylesheets_7.56.0/source/p5subset.xml | perl -pe "s,%,',g;") ; do echo "---------$path:" ; xsel -t -m $path -v "@mode" -n Stylesheets_7.57.0/source/p5subset.xml ; done

Although we may want the Stylesheets to handle this better, the current crisis-fix is to put mode="add" back, reverting #2520. Thus I am moving this ticket to the TEI repo.

Fwiw, I just observed similar weird behaviour in an ODD which redefines @type to use a closed valList. With mode='add' , the schema did the right thing, but the doc did not display the legal values. I changed it to mode="replace" and everything seemed to be OK.

Closed via TEIC/Stylesheets#706