Description of the problem / feature request:

While investigating memory usage during analysis phase noticed that py_binary and py_test rules were responsible for majority of memory usage in analysis phase. Tracked it down to zip action that is always created - irrespective whether --build_python_zip is used, or what platform it runs on.

Looks like the below code iterates over all runfiles and creates strings for those, if using a lot of dependencies this adds on pretty quickly:

Commenting out the above code caused 80% reduction of heap usage in our case as reported by bazel info used-heap-size-after-gc after building codebase with --nobuild option. Even simply calling .intern() on the created String reduced heap usage by 70%, but it maybe very code-base dependent.

Bugs: what's the simplest, easiest way to reproduce this bug? Please provide a minimal example if possible.

See first comment.

What operating system are you running Bazel on?

Ubuntu 20.04

What's the output of bazel info release?

bazel 5