3.7 Some internals on temporary files etc.

It is not totally obvious how TEX processes are supervised by PyX and why it's done that way. However there are good reasons for it and the following description is intended for people wanting and/or needing to understand how temporary files are used by PyX. All others don't need to care.

Each time PyX needs to start a new TEX process, it creates a base file name for temporary files associated with this process. This file name is used as \jobname by TEX. Since TEX does not handle directory names as part of \jobname, the temporary files will be created in the current directory. The PyX developers decided to not change the current directory at all, avoiding all kind of issues with accessing files in the local directory, like for loading graph data, L^ATEX style files etc.

PyX creates a TEX file containing \relax only. It's only use is to set TEXs \jobname. Immediately after processing \relax TEX falls back to stdin to read more commands. PyX than uses stdin and stdout to avoid various buffering issues which would occur when using files (or named pipes). By that PyX can fetch TEX errors as soon as they occur while keeping the TEX process running (i.e. in a waiting state) for further input. The size of the TEX output is also availble immediately without fetching the dvi file created by TEX, since PyX uses some TEX macros to output the extents of the boxes created for the requested texts to stdout immediately. There is a TeX hack --ipc which PyX knows to take advantage of to fetch informations from the dvi file immediately as well, but it's not available on all TEXinstallations. Thus this feature is disabled by default and fetching informations from the dvi is tried to be limited to those cases, where no other option exists. By that TEX usually doesn't need to be started several times.

By default PyX will clean up all temporary files after TEX was stopped. However the usefiles list allows for a renaming of the files from (and to, if existing) the temporary \jobname (+ suffix) handled by PyX. Additionally, since PyX does not write a useful TEX input file in a file and thus a usefiles=["example.tex"] would not contain the code actually passed to TEX, the texdebug feature of the texrunner can be used instead to get a the full input passed to TEX.

In case you need to control the position where the temporary files are created (say, you're working on a read-only directory), the suggested solution is to switch the current directory before starting with text processing in PyX (i.e. an os.chdir at the beginning of your script will do fine). You than just need to take care of specifying full paths when accessing data from your original working directory, but that's intended and necessary for that case.