Fix build metrics report format with long placehold filenames

[davidwendt]

Description

Truncates filenames that appear as multiple placehold_placedhold_... in the Build Metrics Report.
Example show here: https://downloads.rapids.ai/ci/cudf/pull-request/17669/0710ad6/cuda12_x86_64.ninja_log.html (requires VPN).

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[davidwendt]

Here is what it looks like after this fix: https://downloads.rapids.ai/ci/cudf/pull-request/17679/fb12cda/cuda12_x86_64.ninja_log.html (VPN required)

[bdice]

Explaining this behavior for anyone who is curious: Conda uses a "placeholder" intentionally, to ensure that hardcoded paths in binaries are relocatable. On install, conda can replace hardcoded paths in a binary with the proper user path, but it cannot be longer than the original placeholder string because doing so would change the locations of content/symbols in the binary. It can be shorter and stop early with a \0.

I would prefer to use something like this:

import re

def replace_placeholder_patterns(input_string: str) -> str:
    pattern = r'(_h_env_placehold)[_placehold]+'
    return re.sub(pattern, r'\1...', input_string)

print(replace_placeholder_patterns("CMakeFiles/cudf.dir/opt/conda/conda-bld/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeho/include/rmm/logger_impl/logger.cpp.o"))

This gives output like:

CMakeFiles/cudf.dir/opt/conda/conda-bld/_h_env_placehold.../include/rmm/logger_impl/logger.cpp.o

[mhaseeb123]

Looks good. One small non-blocking question though, as format_file_name only returns replace_placeholder_patterns, can we just consolidate them? Would we need to make any other adjustments to the display name in the future to keep it in the current form?

[davidwendt]

Looks good. One small non-blocking question though, as format_file_name only returns replace_placeholder_patterns, can we just consolidate them? Would we need to make any other adjustments to the display name in the future to keep it in the current form?

Right. I expect further updates to this function in the future and this will help keep them in one place.

[vyasr]

Explaining this behavior for anyone who is curious: Conda uses a "placeholder" intentionally, to ensure that hardcoded paths in binaries are relocatable. On install, conda can replace hardcoded paths in a binary with the proper user path, but it cannot be longer than the original placeholder string because doing so would change the locations of content/symbols in the binary. It can be shorter and stop early with a \0.

I would prefer to use something like this:

import re

def replace_placeholder_patterns(input_string: str) -> str:
    pattern = r'(_h_env_placehold)[_placehold]+'
    return re.sub(pattern, r'\1...', input_string)

print(replace_placeholder_patterns("CMakeFiles/cudf.dir/opt/conda/conda-bld/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeho/include/rmm/logger_impl/logger.cpp.o"))

This gives output like:

CMakeFiles/cudf.dir/opt/conda/conda-bld/_h_env_placehold.../include/rmm/logger_impl/logger.cpp.o

JFYI the docs on this are at https://docs.conda.io/projects/conda-build/en/latest/resources/make-relocatable.html.