Fingerprinting

The fingerprinter answers a single question: which features does this tree contain? Features are the keys that drive both template selection and rules-table lookup.

Default detectors

The default Fingerprinter ships with 11 detectors:

Feature	Triggers on
python	pyproject.toml, requirements.txt, setup.py, *.py
node	package.json, *.{js,mjs,cjs,jsx,ts,tsx}
go	go.mod, *.go
rust	Cargo.toml, *.rs
docker	Dockerfile, docker-compose.{yml,yaml}
terraform	*.tf, *.tfvars
jupyter	*.ipynb
java	pom.xml, build.gradle{,.kts}, *.java, *.kt
ruby	Gemfile, Rakefile, *.rb, *.gemspec
csharp	*.{csproj,sln,cs,fsproj,vbproj}
swift	Package.swift, *.swift

A 12th feature, common, exists only as a template (.DS_Store, .idea/, .vscode/, Thumbs.db, etc.). It is not detected — it is implicitly included by generate() whenever the fingerprint is non-empty and a common template is present in the repository.

This decoupling matters: the fingerprint stays a pure description of what the tree contains. The decision to add OS/IDE patterns is a generation policy, applied in the generator.

Result shape

@dataclass(frozen=True, slots=True)
class FingerprintResult:
    features: tuple[Feature, ...]      # sorted, deduplicated
    evidence: tuple[tuple[str, str], ...]  # sorted (feature_name, witness_path)

evidence carries one witness path per detected feature — useful for debugging and explainable output.

Extending detectors

Add a Detector and pass a custom tuple to DefaultFingerprinter(...):

from occam_gitignore_core import DefaultFingerprinter, Detector, Feature

custom = (
    *DefaultFingerprinter._DETECTORS,  # type: ignore[attr-defined]
    Detector(Feature("haskell"), lambda p: p.endswith((".hs", ".cabal"))),
)
fp = DefaultFingerprinter(detectors=custom)

In practice, prefer adding the detector to fingerprint.py and submitting a PR — that way it benefits everyone, the rules-table miner sees the new feature, and the bench corpus can cover it.