This project seems interesting: http://upx.sourceforge.net/
This could potentially reduce the size of our wheels significantly to both reduce bandwidth usage and speed up install times with pip. We could add an option to auditwheel repair to enable UPX compaction on the compiled extensions and the grafted libraries. Another benefit would be to reduce on-disk size of the typical virtualenv holding the full scipy stack. This can be important to deploy it in serverless environments such as AWS lambda which has size restrictions (50MB) for the hosted code. For instance see this blog post on this topic (it does not use UPX, only stripped binaries): https://serverlesscode.com/post/deploy-scikitlearn-on-lamba/ Note that this second use case (venv compaction) could easily be addressed by a tool not related to wheels but I thought that using UPX upstream in the build & packaging process might be of general interest (primarily to save bandwidth and speedup installs). -- Olivier http://twitter.com/ogrisel - http://github.com/ogrisel _______________________________________________ Wheel-builders mailing list [email protected] https://mail.python.org/mailman/listinfo/wheel-builders
