Ignore files with a UTF-8 BOM: first pattern silently dropped (git honors it, graphify does not) #2163

Affected version

• graphifyy 0.9.25 (uv tool install)
• Windows 11, PowerShell 5.1 + Git for Windows

Summary

.gitignore and .graphifyignore are parsed without stripping a leading UTF-8 BOM. When the file is BOM-encoded, the first pattern is silently dropped — it is read as <BOM><pattern>, which matches nothing.

Git itself strips the BOM, so a file git honors is silently ignored by graphify. There is no warning.

This is easy to hit on Windows: PowerShell 5.1's Set-Content -Encoding utf8 — the most obvious way to write the file — emits a BOM.

Reproduce

Fixture: a repo with keep/a.py plus five .py files under .fable-wt/wt1/.

# BOM-free
[System.IO.File]::WriteAllText("$t\.graphifyignore", ".fable-wt/`n", (New-Object System.Text.UTF8Encoding($false)))
graphify . --code-only     # -> found 1 code    CORRECT

# identical content, with BOM
[System.IO.File]::WriteAllText("$t\.graphifyignore", ".fable-wt/`n", (New-Object System.Text.UTF8Encoding($true)))
graphify . --code-only     # -> found 6 code    PATTERN SILENTLY IGNORED

# what a Windows user actually types
Set-Content -Path "$t\.graphifyignore" -Value ".fable-wt/" -Encoding utf8
graphify . --code-only     # -> found 6 code    PATTERN SILENTLY IGNORED

Measured:

Divergence from git

Same pattern, same BOM, in .gitignore:

git honors BOM'd .gitignore:     True
git honors non-BOM .gitignore:   True
graphify w/ BOM'd .gitignore:    1 code file found (expected 0)

graphify documents that it honors .gitignore, but diverges from git on BOM-encoded files.

Impact

Silent. On a monorepo where the intended exclusion is a worktree root, dropping that one pattern is the difference between indexing one copy of the tree and indexing 100+. In my case the first pattern was .fable-wt/ against 111 worktrees; the run went from seconds to 27+ minutes of CPU with no indication the ignore file was not in effect.

Possibly related: #1958 (detect() CPU-bound 35+ min on large monorepo).

Suggested fix

Read ignore files with utf-8-sig, which transparently handles both BOM and BOM-free input. In detect.py, in the per-directory reader and _load_graphifyignore:

text = path.read_text(encoding="utf-8-sig")

Optionally also warn when an ignore file is BOM-encoded, since the failure is otherwise invisible.