Ulrike Fischer a écrit :
Why guessing? And why should convert luatex from utf8 to latin1 and back? If luatex opens a file it can check if the file contains a line about the encoding and then apply this encoding.
Because to convert the main tex file, I need to put a function in the callback process_input_buffer, and it converts everything from latin1 to utf, so if I want to have a file in utf, I need to convert it first (before it is processed in process_input_buffer) in latin1, and then (in process_input_buffer) convert it again to utf. If there is another solution than the process_input_buffer to convert the main tex file, I take it though...
Above you are complaining that too much converting is involved and now you doing unnessary converting yourself. In your scenario you are converting the internal utf8 to latin1 to write the aux, and you are converting the content of the aux again to utf8 to handle it during compilation. In my scenario you only write a comment to the aux so that luatex knows that it is already an utf8-file and can be handle directly -- no input converting and no output converting needed. But if you really want to get aux-files in latin1: You don't need luatex to be able to write directly in latin1. You only need to be able to convert an utf8-file in e.g. latin1. You could do it after you closed the file.
I might have found a better solution that (unexpectedly) works: calling inputenc after lua-inputenc... it works fine for my tests: it uses the LICR, so when it reads the file again, no problem, but I need to test it more... If it works, I'll take this solution, as my aim is to emulate the old inputenc, not creating a new one with new aux files encoding conventions. I think people who want to do new things with LuaTeX must be workin in UTF-8, not using my packet (that will work for them too anyway).
-- Elie
