ZJU-SPAIL / pipa / 17755914047

70%

Build Type

push

github

Committed by bernard035

Commit Message

fix(pipashu): resolve initialization failures and
refactor parsers

This commit addresses critical bugs in the
`PIPAShuData` initialization process, ensuring it
 runs successfully on modern systems with newer
`perf` versions.

The following fixes and refactors are included:

1.  **Resolved `multiprocessing` Serialization
Error**:
    - Replaced a non-serializable lambda function
     in `PerfScriptData.from_file` with a static
     helper method to ensure compatibility with
     `multiprocessing`.

2.  **Refactored `PerfStatParser` for
Robustness**:
    - The original `PerfStatParser` was fragile
    and failed on newer `perf` versions. It has
     been completely refactored to use a robust
      "pivot-and-melt" algorithm.
    - The new parser correctly handles varying
    column counts, different event orders, and
     maintains backward compatibility.

3.  **Fixed Logic in `PerfStatDataProcessor`**:
    - Corrected the event name matching in
     `get_CPI` and `get_events_overall` from
     exact match (`==`) to prefix match
     (`startswith`) to handle event suffixes
      like `:D`.

4.  **Added Comprehensive Unit Tests**:
    - A new test suite (`test_perf_stat.py`) has
    been added to cover the refactored parser and
     data processor.
    - Tests cover modern, legacy, and aggregated
    `perf stat` formats, as well as edge cases
     like empty and malformed files.
    - This significantly improves the code
    coverage and reliability of the `perf_stat`
     module.

5.  **Incorporated robustness suggestions from
    Copilot review.**
    - Added checks to prevent accessing DataFrame
    rows when empty, avoiding potential runtime
     errors.

These changes collectively make the core data
loading functionality of PIPA significantly more
 robust and reliable, allowing the Quickstart
 guide to be completed successfully.

Run Details

98 of 148 new or added lines in 3 files covered. (66.22%)

1 existing line in 1 file now uncovered.

378 of 2534 relevant lines covered (14.92%)

0.15 hits per line