This speeds up the latency process, even though it increases the total work (because before_all.sh now runs for every wheel, instead of once per architecture).
A new stage is added at the end to fetch the debug symbols produced for each wheel and put them into a single compressed tarball. This gets much better compression than the zipfile that Github produces for the artifact (I'm guessing because the individual files have a lot of commonality, and a .tar.xz is a compressed archive of files while .zip is an archive of compressed files.
This speeds up the latency process, even though it increases the total work (because before_all.sh now runs for every wheel, instead of once per architecture).
A new stage is added at the end to fetch the debug symbols produced for each wheel and put them into a single compressed tarball. This gets much better compression than the zipfile that Github produces for the artifact (I'm guessing because the individual files have a lot of commonality, and a .tar.xz is a compressed archive of files while .zip is an archive of compressed files.