differens-correlate

Find declaration and whole-file moves between per-file edit lists.

Main exports

Rust API	Purpose
`correlate`, `FileChanges`, `CorrelateOptions`	Correlate an in-memory list of per-file edits
`CrossFileMatch`, `ActionId`	Inspect the pair and its original file/action positions
`is_move_candidate`	Check candidate eligibility
`CandidateStore`, `correlate_store`	Supply indexed candidate storage for larger changesets

Options and results

use differens_correlate::{correlate, FileChanges, CorrelateOptions};

fn moves(files: &[FileChanges]) -> Vec<differens_correlate::CrossFileMatch> {
    correlate(files, CorrelateOptions {
        rename_similarity_threshold: 0.6,
    })
}

FileChanges contains file_path: String and actions: Vec<EditAction>. correlate(&[FileChanges], CorrelateOptions) -> Vec<CrossFileMatch> preserves input order and consumes each source/destination action at most once. A match includes node, from_file, to_file, modified, similarity, and stable ActionId { file, action } values for deletion and insertion. IDs refer to the original ordered inputs; the correlator does not mutate those lists. is_move_candidate(&EditAction) accepts named declaration and whole-file insertions/deletions. It rejects generic syntax fragments, updates, existing moves, and local variables inside named declarations. An unlabeled export wrapper can expose its named child. Exact matching verifies full content/topology behind every hash match. Whole-file renames can match identical contents despite different path labels. Modified matching uses token-set Jaccard similarity, with the default threshold 0.6. Declaration names must agree for modified moves; whole-file path names do not affect similarity. Structure buckets exceeding 64 candidates are filtered by name and skipped if still ambiguous. A similarity match is a heuristic. Unmatched edits remain ordinary insertions/deletions. Reporting a move does not prove a code extraction or a behavioral equivalence.

Correlate real per-file diffs

Keep unchanged declarations in both files so each source is parsed as an existing file. The function moves from utils.ts into an existing pricing.ts:

use differens_correlate::{CorrelateOptions, FileChanges, correlate};
use differens_tiers::{SourcePresence, diff_with_tier};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let function = "export function formatPrice(price: number) { return price.toFixed(2); }\n";
    let inputs = [
        ("utils.ts", format!("{function}export const TAX_RATE = 0.08;\n"),
                     "export const TAX_RATE = 0.08;\n".to_owned()),
        ("pricing.ts", "export const CURRENCY = \"USD\";\n".to_owned(),
                       format!("export const CURRENCY = \"USD\";\n{function}")),
    ];
    let files: Vec<_> = inputs.into_iter().map(|(path, old, new)| {
        let diff = diff_with_tier(old.as_bytes(), new.as_bytes(), path, path, SourcePresence::default());
        FileChanges { file_path: path.to_owned(), actions: diff.changes }
    }).collect();
    let moves = correlate(&files, CorrelateOptions::default());
    for movement in moves {
        println!("{} {} -> {} edited={} similarity={:.2}",
            movement.node.label.as_deref().unwrap_or(""),
            movement.from_file, movement.to_file, movement.modified, movement.similarity);
    }
    Ok(())
}

formatPrice utils.ts -> pricing.ts edited=false similarity=1.00

rename_similarity_threshold defaults to 0.6 and must be finite and in 0.0..=1.0; invalid options produce no matches. Exact matches use similarity 1.0 and modified = false. A similarity match sets modified = true, even if the token sets happen to score 1.0 after some non-token edit. Whole-file matches are reported as file renames. For a complete CLI example, including a moved function edited in transit, see the cross-file walkthrough.