Update PythonZipper action to use CommandLineItem.CapturingMapFn

[philsc]

Summary

This PR attempts to address #14890.

This updates the PythonZipper action to use a CommandLineItem.CapturingMapFn to defer expanding arguments until after analysis. I used CapturingMapFn, as it seemed like the only way to go about implementing the arg filtering that is done by the current code to exclude the executable and zipFile artifacts.

Test Plan

Use the example from #14890 of looking at memory usage in tensorflow before/after this patch is applied on top of version 5.0.0

Initial Setup

git clone [email protected]:tensorflow/tensorflow.git
cd tensorflow
python3 -m venv venv
source venv/bin/activate
pip install numpy

View memory usage at 5.0.0:

# STARTUP_FLAGS from https://bazel.build/rules/performance#enabling-memory-tracking
$ bazel ${STARTUP_FLAGS[@]} build --nobuild //tensorflow/python/...
$ bazel ${STARTUP_FLAGS[@]} dump --rules | egrep '(RULE|py_)'
RULE                                 COUNT     ACTIONS          BYTES         EACH
py_test                              1,714      15,390    413,338,632      241,154
py_library                           1,102           0      2,097,152        1,903
py_binary                               33         198      8,914,840      270,146
py_runtime                               6           0              0            0
py_runtime_pair                          3           0              0            0
_concat_flatbuffer_py_srcs               2           2              0            0

$ bazel ${STARTUP_FLAGS[@]} info used-heap-size-after-gc
635MB

View memory usage at 5.0.0 with this patch applied:

$ ~/code/bazel/bazel-bin/src/bazel ${STARTUP_FLAGS[@]} dump --rules | egrep '(RULE|py_)'
RULE                                 COUNT     ACTIONS          BYTES         EACH
py_test                              1,714      15,390     65,323,312       38,111
py_library                           1,102           0      2,359,576        2,141
py_binary                               33         198        524,400       15,890
py_runtime                               6           0              0            0
py_runtime_pair                          3           0              0            0
_concat_flatbuffer_py_srcs               2           2              0            0

$ ~/code/bazel/bazel-bin/src/bazel ${STARTUP_FLAGS[@]} info used-heap-size-after-gc
283MB

Ensure that the generated actions have not changed:

$ bazel aquery --build_python_zip 'mnemonic("PythonZipper", //tensorflow/python/...)' > /tmp/bazel_5_out
$ ~/code/bazel/bazel-bin/src/bazel aquery --build_python_zip 'mnemonic("PythonZipper", //tensorflow/python/...)' > /tmp/bazel_5_patched_out
$ diff /tmp/bazel_5_out /tmp/bazel_5_patched_out
# note: no diff output

(cherry picked from commit 9610ae8)

[philsc]

I have no clue what that windows failure is about. The error message is not meaningful to me.

[ckolli5]

@philsc Could you please rebase your branch with the latest release-5.2.0 branch? Thanks!

[philsc]

Done