Gh1359

[FrankMittelbach]

Internal housekeeping

Status of pull request

• Ready to merge

Checklist of required changes before merge will be approved

• [x] Test file(s) added
• [x] Version and date string updated in changed source files
• [x] Relevant \changes entries in source included
• [x] Relevant changes.txt updated
• [n/a] Rollback provided (if necessary)?
• [x] ltnewsX.tex (and/or latexchanges.tex) updated

This is a straight bug in code that hasn't been used that much yet, so I think a rollback isn't really needed.

I don't understand one test file change: github-1041, why that goes from no room for insert to no room for count in luaTeX, maybe @davidcarlisle can verify that as it is a test for code written by him.

Otherwise I think this would be ready to go.

[josephwright]

I took a look at the failure in github-1041 and I think I see what's happening. With LuaTeX, we can create inserts in the extended pool, which means that the code path deviates from the classical one. Doing that, we call \extrafloats, and that does some allocation. What's happening I think is the extra count being consumed is meaning that rather than hit the buffers in \newinsert, we take a path in \extrafloats that tries to allocate just a count - and that now fails.

The code for allocators is all a bit baroque, not least because there are a lot of conditionals for pre-e-TeX, etc. I think for the present we live with the change of .tlg here, but perhaps should look to re-work the allocator code a bit more widely to make the logic clearer. (Not exactly a high priority, but would help with this sort of edge case thing.)