luaotfload-unicode.lua fails when near UnicodeData.txt

[nxg]

The program luaotfload-tool fails when a file UnicodeData.txt is in the current directory.

% luaotfload-tool --update
% echo hello >UnicodeData.txt
% luaotfload-tool --update
.../texmf-dist/tex/luatex/luaotfload/luaotfload-unicode.lua:108: assertion failed!
stack traceback:
[C]: in function 'assert'
.../texmf-dist/tex/luatex/luaotfload/luaotfload-unicode.lua:108: in main chunk
[C]: in function 'require'
...texmf-dist/tex/luatex/luaotfload/luaotfload-database.lua:202: in main chunk
[C]: in function 'require'
/Library/TeX/texbin/luaotfload-tool:167: in local 'loadmodule'
/Library/TeX/texbin/luaotfload-tool:188: in main chunk

Running lualatex also fails in this situation. Removing the file UnicodeData.txt from the current directory is enough to make both programs work again. The above is with a dummy file; it fails in exactly the same way with the real UCD file there.

The lines 107--109 in question are

local f = io.open(kpse.find_file"UnicodeData.txt")
assert(file:match(f:read'*a'))
f:close()

I appreciate this is unlikely to affect many people, and the workaround I can apply is obvious, but this is presumably not quite working as designed.

[davidcarlisle]

I only get the assert failure if the files don't match:

davidc@david-14:~$ echo hello > UnicodeData.txt
davidc@david-14:~$ kpsewhich --all UnicodeData.txt
./UnicodeData.txt
/usr/local/texlive/2026/texmf-dist/tex/generic/unicode-data/UnicodeData.txt
davidc@david-14:~$ luaotfload-tool --update
.../texmf-dist/tex/luatex/luaotfload/luaotfload-unicode.lua:108: assertion failed!
stack traceback:
        [C]: in function 'assert'
        .../texmf-dist/tex/luatex/luaotfload/luaotfload-unicode.lua:108: in main chunk
        [C]: in function 'require'
        ...texmf-dist/tex/luatex/luaotfload/luaotfload-database.lua:202: in main chunk
        [C]: in function 'require'
        /usr/local/texlive/2026/bin/x86_64-linux/luaotfload-tool:167: in local 'loadmodule'
        /usr/local/texlive/2026/bin/x86_64-linux/luaotfload-tool:188: in main chunk


davidc@david-14:~$ rm UnicodeData.txt
davidc@david-14:~$ cp `kpsewhich UnicodeData.txt ` .
davidc@david-14:~$ luaotfload-tool --update
davidc@david-14:~$

shows the failure you show with hello but no failure if the local file is a copy of the installed one.

The assert is checking that the content of UnicodeData.txt is as expected, so I think the failure is "by design" (although it could have a friendlier message) But I'll leave for @zauguin to comment on that.

[zauguin]

This is mainly by design. We are following regular kpathsea conventions and I am reluctant to deviate from that since it would be inconsistent with LuaLaTeX runtime behavior.
Also especially for UnicodeData.txt I quite like being able to run easily with a different version for testin ;) But this might change soon-ish since we consider switching to more pre-computed data, so we might end up with less generic names being read.

Also the requirement isn't so much that it needs to be the same, but I needs to match the structure of a Unicode Character Database data file.

We can improve the error message here a bit, but the message would probably be along the lines of "Unicode database file corrupted" which still doesn't make it super clear that a local file is the issue, but having a UnicodeData.txt lying around is not really common enough to add the complexity of handling that case separately.