bug(indexer): Non-UTF-8 source files are silently dropped for Python, JS, TS, TSX, Go, CSS, HTML #1390

tools/languages/python.py:150, javascript.py:187, typescript.py:181, typescriptjsx.py:84, go.py:180, css.py:20, html.py:42

with open(path, "r", encoding="utf-8") as f:      # no errors= parameter
    source_code = f.read()

The other 20 language parsers (c, cpp, java, kotlin, rust, php, ruby, swift, scala, dart, perl, csharp, haskell, elixir, elisp, lua, …) all pass errors="ignore". The seven most-used languages do not.

UnicodeDecodeError is caught by graph_builder.py:1053, which returns {"path": …, "error": …} with no unsupported key. pipeline.py:149-155 then treats it as a generic file and creates a bare File node — indistinguishable from a .md file.

Reproduced. A repo with latin.py (latin-1 ç), ok.py, bom.py:

Total scanned files: 3 | Function nodes: 4
Successfully finished indexing

Graph:  File node for latin.py       → exists
        Function nodes for latin.py  → NONE

No warning; the summary table has no "files failed" row. A codebase with legacy latin-1 files loses them entirely and reports success.

Fix. Use errors="replace" consistently across all 23 parsers, and add a failed_files count to build_index_summary.

_{Audit ID I-H02 · severity high · from an end-to-end audit of CGC 0.5.2 (63 findings). Reported with the reproduction as run.}