MCPcopy Create free account
hub / github.com/atomicdotdev/atomic / globalize_replace

Function globalize_replace

atomic-core/src/record/workflow/globalize/hunk.rs:313–492  ·  view source on GitHub ↗
(
    ctx: &mut GlobalizeContext<'_, T>,
    built: &BuiltHunk,
    inode: Inode,
    inode_pos: Position<NodeId>,
    content: &[u8],
    full_content: &[u8],
    local: Local,
    encoding: Option<E

Source from the content-addressed store, hash-verified

311/// before per-line vertex creation), we fall back to whole-file Replace.
312#[allow(clippy::too_many_arguments)]
313fn globalize_replace<T>(
314 ctx: &mut GlobalizeContext<'_, T>,
315 built: &BuiltHunk,
316 inode: Inode,
317 inode_pos: Position<NodeId>,
318 content: &[u8],
319 full_content: &[u8],
320 local: Local,
321 encoding: Option<Encoding>,
322) -> GlobalizeResult<Vec<GraphOp<Option<Hash>>>>
323where
324 T: GraphTxnT + TreeTxnT + InodeGraphOps,
325{
326 // NOTE (review ::26 R1 follow-up, 2026-09-16): the historical
327 // `should_use_opaque_generated_vertices` early-return here routed EVERY
328 // Replace hunk of an opaque path (lockfiles/checksums) to whole-file
329 // replacement — using the hunk's SLICED content (just the modified
330 // lines). A lockfile-only modification of lines 2 and 6 therefore
331 // became two sequential whole-file replaces carrying one line each,
332 // destroying the other lines (12-of-67 bytes rendered — a real
333 // data-loss defect). Targeted per-line surgery is classification-blind:
334 // the deletion and insertion operate on graph vertices, so an
335 // opaque-classified file with per-line structure takes the SAME
336 // targeted path. The monolithic-vertex and empty-file shapes still
337 // fall back to whole-file replacement below (bounds-check +
338 // sorted.is_empty()), and whole-file lifecycle paths (add/delete/
339 // globalize_replace_whole_file) keep their opaque insert shape.
340
341 // Every whole-file fallback below rewrites the file from the FULL new
342 // content: the hunk slice holds only the replaced lines, and a replace
343 // built from it deleted every vertex and kept just those lines (a
344 // one-line tox.ini edit on a file whose vertices no longer matched its
345 // lines rendered 63 of 855 bytes and the import refused).
346 let sorted = collect_sorted_content_vertices_cached(ctx, inode, inode_pos)?;
347
348 // Legacy: file has no visible content vertices in this view. Fall
349 // back to whole-file replace, which walks INODE_GRAPH unfiltered.
350 //
351 // NOTE: we use `== 0`, not `<= 1`. A single-line file produces
352 // exactly one per-line vertex, and the targeted path below handles
353 // that just fine (predecessor=inode, successor=None, delete that
354 // one vertex, insert the replacement lines). Falling back to
355 // whole-file replace for single-line files would have us re-read
356 // edges through the unfiltered INODE_GRAPH index, which picks up
357 // vertices from OTHER views and silently adds their changes as
358 // dependencies — producing a phantom supersession relationship
359 // between the just-recorded change and the other view's edits.
360 if sorted.is_empty() {
361 return globalize_replace_whole_file(ctx, inode, inode_pos, full_content, local, encoding);
362 }
363
364 // Pure insertion (no lines deleted) — treat as a middle/append insert.
365 // This happens when the diff produces a Replace with old_len == 0
366 // (an insertion that the algorithm classified as a replacement).
367 if built.deleted_lines.is_empty() {
368 let position =
369 match classify_insert(&sorted, inode_pos, built.old_start, Some(sorted.len())) {
370 Ok(p) => p,

Callers 1

globalize_hunkFunction · 0.85

Calls 15

classify_insertFunction · 0.85
InsertionClass · 0.85
build_deletion_edgesFunction · 0.85
EdgeUpdateClass · 0.85
lastMethod · 0.80
is_emptyMethod · 0.45
lenMethod · 0.45
collectMethod · 0.45

Tested by

no test coverage detected