On Mon, Oct 30, 2023 at 05:45:28PM -0400, Matt Welland wrote: > Well, maybe there is a silver lining here, I can currently reliably > reproduce the problem!
That's excellent! Do you have to do any special steps, or does it simply always happen? > One hypothesis I'm looking at is replacing calls using with-input-from-file > and with-output-to-file with open-*-file ... close loops for the reading > and creation of the egg-info files. You're probably on the wrong track there. AFAICT, egg-info files are only *created* in the foo.install.sh / foo.install.bat file, at the very end where (on *nix) it uses cat(1) to write into it. If the file is already empty, changing how it's read won't make a difference. The only way it would fail that I can see is if one presses ^C to abort installation at a very unlucky time. This sounds like a race condition but depending on the shell, it might be that it will process that signal after evaluating the line that redirects stdout to the egg-info file, but before actually invoking cat. Dunno, grasping at straws here :) Since you're using mingw, it's a UNIX shell at least. Could you try inserting a sync(1) call at the end of the .install.sh file to see if that fixes things? Cheers, Peter
signature.asc
Description: PGP signature