Simplify and speed up GDSII diffs

Summarize added and removed cells, group references, and keep replacement pairs together at text display limits. Trim hierarchy labels and summarize long unmatched paths.

Stream JSON object evidence and reduce parsing and comparison allocations. Parse large revisions concurrently and adaptively reuse matching cell bodies when most of the layout is shared.
This commit is contained in:
Jan Petykiewicz 2026-11-18 10:58:19 -08:00
commit 8efeb04aab

View file

@ -26,10 +26,19 @@ Layout-review APIs live in `pheon::review`, with layer-name helpers in
`Comparison::between(&old, &new)` captures an owned result with revision metadata,
changes, and structured diagnostics. Use `outcome()` for completeness, equality,
and exit status. Its `text`, `json_report`, and `hierarchy` renderers work after
the source documents and byte buffers are dropped. Standalone hierarchy inspection
uses `Document::hierarchy`. Reference counts are captured with the result;
union-graph and ancestor analysis run only when a hierarchy is requested.
and exit status. Its renderers work after the source documents and byte buffers
are dropped. `json_report` returns a JSON value, and `write_json_report` writes
object evidence incrementally to a writer. `Document::hierarchy` inspects a
single revision. A comparison captures reference counts and analyzes the union
graph and ancestors when a hierarchy is requested.
`TextComparison::between` captures the data needed for text and hierarchy
reports. Layered objects in added and removed cells contribute to the cell's
object count.
Use `RevisionPair::parse(&old_bytes, &new_bytes)` for an unscoped comparison
that can reuse matching cells. Its `compare` and `compare_text` methods return
the corresponding owned comparison.
The `record`, `records`, and `basic` modules also provide low-level GDSII record
writing. There is no high-level writer for the Arrow layout. REAL8 writing is