Documentation
¶
Overview ¶
Package diffengine ports src/pr_af/diff_engine.py — the deterministic (non-LLM) diff parser and clustering engine that feeds the anatomy phase.
It is a byte-for-byte behavioral port of the Python module. Where the Python implementation has quirks (see notes below) they are reproduced exactly rather than "fixed", because downstream golden comparisons and cluster/finding ordering depend on identical output:
- parse_unified_diff never emits a "renamed" status. Git rename headers (`rename from`/`rename to`, `similarity index`) are not special-cased; a rename surfaces as status "modified" with the path taken from the `+++ b/` line (or "" for a pure content-free rename that has no `+++ b/`).
- A deleted file (`+++ /dev/null`) never sets a path, so its Path is "".
- Binary files (no `---`/`+++` lines) yield a FileChange with Path "" and no hunks.
- Line splitting matches Python's str.splitlines() (not strings.Split on "\n"): all Unicode line boundaries split, and a trailing line break does NOT produce a final empty element. This keeps hunk `content` byte-exact.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func ClusterChanges ¶
func ClusterChanges(files []schemas.FileChange) []schemas.ChangeCluster
ClusterChanges ports cluster_changes: group files by their directory (the path up to the last "/"; "root" when there is no "/"), then emit one cluster per directory in sorted-directory order with deterministic ids "cluster_0", "cluster_1", ... The primary language is the most frequent language in the group.
Divergence note: Python computes primary_language via `max(set(langs), key=langs.count)`, whose tie-break depends on set iteration order and is therefore non-deterministic across Python processes (str hash randomization). This port instead breaks ties deterministically by first-appearance order within the group. For a unique mode (the overwhelming common case, including single-language groups) both agree exactly.
func ComputeDiffStats ¶
func ComputeDiffStats(files []schemas.FileChange) schemas.DiffStats
ComputeDiffStats ports compute_diff_stats: aggregate statistics over the parsed files. The test-to-code ratio divides the test-file count by max(code_files, 1) so it never divides by zero.
func ParseUnifiedDiff ¶
func ParseUnifiedDiff(diffText string) []schemas.FileChange
ParseUnifiedDiff parses unified diff text into structured FileChange objects. It is a direct port of Python parse_unified_diff, preserving branch order and the accumulation quirks documented on the package.
Types ¶
This section is empty.
Source Files