Bug Tracker

I forgot to mention, but I tried to play around with mitigations, and found that it seems to work again if I do something like:

\ExplSyntaxOn
\renewcommand{\glslowercase}[1]{\edef\tmp@gls@lowercase@arg{#1}\text_lowercase:n{\tmp@gls@lowercase@arg}}
\renewcommand{\glsuppercase}[1]{\edef\tmp@gls@uppercase@arg{#1}\text_uppercase:n{\tmp@gls@uppercase@arg}}
\ExplSyntaxOff

A better fix is to use \exp_args:Ne:

\renewcommand\glsuppercase[1]{\exp_args:Ne \text_uppercase:n { #1 } }

Thank you.

Thanks, Nicola!

I did a little more digging today, and I think I've found out why the expansion logic changed. Initially, I was wondering if this could be a bug in the l3kernel (if the change in behavior was not intended). However, now it looks like the change was very much intended!

Right now, I've been focusing specifically on \glsentrytext and why \glsuppercase{\glsentrytext{dll}} did not work anymore. I managed to trace it down to this: \glsmfuaddmap{\glsentrytext}{\Glsentrytext} which to my understanding (but there is a lot I don't understand) ends up doing something similar to \glsmfuexcl{\glsentrytext}, which in turn seems to add \glsentrytext to \l_text_case_exclude_arg_tl.

Now, consider this change: [TUG Link] which was part of the revision I linked to in the original bug description. I found the corresponding change on the latex3 GitHub: [Link], and it seems it was an intended change to fix another bug.

I don't really understand what \l_text_case_exclude_arg_tl is, and certainly don't understand enough about your glossaries packages to be able to suggest a fix. But, as a software developer, I understand enough to be able to assess that the change (although intended) has caused a regression in your packages that needs fixing, and I thought I'd share this information because it seems to be important as part of triaging the issue to find the "right" fix.

Hi, thank you for the extra information.

The exclusion list (\l_text_case_exclude_arg_tl) is to prevent labels or similar arguments from having a case-change. For example, \ref is automatically added as an exclusion by the L3 kernel to prevent \text_uppercase:n { see ~ \ref{somelabel} } from being converted to SEE ~ \ref{SOMELABEL}. The robust commands like \gls are added as exclusions by the glossaries package. The mapping from \glsentrytext to \Glsentrytext ensures that \makefirstuc{\glsentrytext{label}} is converted to \Glsentrylabel{label} but this automatically adds the mapped command (\glsentrytext) as an exclusion.

I think that since \glsentrytext is expandable the map can probably be removed. It's only needed for the robust commands (such as \gls which is mapped to \Gls).

I've had another look at the LaTeX3 manual and the case-changing commands in the l3text module are still described as having their argument expanded using \text_expand:n so I think that the problem is due to the mfirstuc mappings which have the side-effect of adding the \glsentry... commands as exclusions. The issue seems to resolve itself by changing \glsmfuaddmap{\glsentrytext}{\Glsentrytext} to \glsmfuexcl{\Glsentrytext} (and similarly for the other \glsentry... commands).

Brilliant, thank you for producing a fix! :-)