mirror of
https://github.com/penpot/penpot.git
synced 2026-10-29 14:00:37 +00:00
69a4ef273d
The comparison pairs the branch's media rows with main's through the
content identity heuristic (media-pairs), which approximates the
pairing from the rows present on both files. The writes that actually
establish the pairing know it exactly, so now they record it and the
comparison reads the record.
Every file_branch_change row stamps media_id_map jsonb: the media id
map the save or the update from main that wrote the row applied, as
applied ({old-id -> new-id}, the branch's own row is the new one). A
save stamps the remap its media fix-up applied, filtered to the copies
whose source row the branch's source file owns; the squash row of an
update from main stamps the map applied to main's data before the
squash (the pairs plus the fresh copies).
files_branch.clj::branch-media-pairs unions the inverted stamps when
the log is fully stamped and falls back to the heuristic when any row
carries none (rows written before the column existed), so old logs
keep working and the heuristic never runs for stamped ones. Every
former media-pairs consumer goes through it: the comparison
(branch-id-map), both base normalizations, the merge and the update.
Signed-off-by: Alvaro Tejero Cantero <alvorithm@teje.ro>
AI-assisted-by: opus-5-5