On 24/03/2015 22:32, Karl Berry wrote: > Yes. (We could happily make ltunicode loadable by plain, but that might > not be that helpful.) > > I was wondering about that. It actually sounds like a good thing in > terms of maintenance to me. I would make a physical copy inside TL so > that all of LaTeX is not pulled in for it, but having only one file, > generated in one way, sounds preferable, if it's easy enough to do. I will make the (minor) changes later today (basically, I did a few things the 'LaTeX way' for the file that I'll simply drop back to the 'primitive way'). One question this raises is the file name. Files that can be loaded by plain or LaTeX are usually .tex or .def, and perhaps "lt..." is wrong here too. I guess replacing unicode-letters.tex with a team-maintained file of the same name is not a great plan, so would suggest something like unicode.def (or perhaps unicode-letters.def): thoughts? (I'll change the code today but not the name unless there is a clear feeling one what makes sense.) > Both EastAsianWidth.txt and LineBreak.txt have a version which we've > talked about copying in to the processed file. Regrettably, there is > no version in UnicodeData.txt: > > Yes, that's why I suggested something else :). I think it's also quite > possible that Unicode has released ("updated") differing files with the > same version number. Unicode versions in general are a hard problem. I thought briefly about using the MD5 sum as that could be done within the script when using pdfTeX. However, that seems harder for a human to check than simply having a line in the log 'Remember to edit in the file sizes'. An alternative (or addition) might be to count the total number of lines parsed. -- Joseph Wright