[MartinThoma]

I just stumbled over this on pdfly: py-pdf/pdfly#295

I still need to test/benchmark this change.

General Problem Analysis

The for-loop with retval += ceffectively needs to re-create and copy the string again and again in the worst case.

pdfly problem analysis: Why only on Windows?

According to Mistral AI, this is an Issue only on Windows with Python 3.11, because glibc (realloc -> mremap) is effectively linear but Windows UCRT (realloc always copies) is quadratic. Anything tracing each executed line amplifies it further, e.g. covrage.py on Python < 3.12 turned a 0.3s clone into 66s in CI

Benchmarks to be done

• columns: current main, PR PI: Decode PDFDocEncoding strings with a translation instead of per-character concatenation #4429
• rows: ASCII string, Linux, py3.15; ASII string, Linux, py3.11.9; non-ASCII PDFDoc text; PdfWriter(clone_from) of a 2.3 MB annotator PDF under coverage.py py3.11, ...

Proof of functional eqivalence

TODO

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.87%. Comparing base (1e3e7e3) to head (11ecb15).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4429   +/-   ##
=======================================
  Coverage   98.87%   98.87%           
=======================================
  Files          59       59           
  Lines       12299    12299           
  Branches     2342     2341    -1     
=======================================
  Hits        12161    12161           
  Misses         68       68           
  Partials       70       70           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.