mirror of
https://github.com/stablyai/orca.git
synced 2026-10-06 16:02:25 +00:00
Jinwoo-H
b78519f946
C7.1 landed on main as a squash, so the merge base predates it and every file C7.1 touched conflicted with this branch's copy of the same work. Sixteen of the eighteen were byte-identical on main to C7.1 at0ce0fc99a2, which this branch already merged, so its side — C7.1's content plus this lane's — was taken whole. The squash also resurrected five files this branch had deleted, since to git they are files main added rather than files this branch removed: `terminal-document-flip.test.ts`, its `terminal-document-pre-flip-script.txt` fixture, and the comparator, its token reader and its test. All five are deleted again, for the reasons their own commits gave. Two files needed a real resolution. `build-mobile-web-app-bundle.mjs` takes main's version whole — including the sentence naming `MobileBrowserPane`'s pin, which now exists, and the Zod jitless banner — with this branch's `absWorkingDir` parameter re-applied on top. `web-overrides.json` keeps both lists: 24 entries here and 28 there, 30 in the union, no duplicates and nothing dropped from either side. Two measurements moved because main moved, and both are now stronger. The render check's CSP control read `script-src: eval` before this merge — Zod probing for a JIT with `new Function` and swallowing the throw, which I reported as pre-existing. Main's jitless banner closed it, so the control now records nothing at all. The subtraction stays: it is what makes the terminal's zero mean "added none" rather than "none were seen". The session closure's 16 px census pinned nine offenders. Main carries C7.2, which moved all nine onto the seam, so the answer is none — and because zero is also what a walk that read nothing reports, the case now names two preconditions: the seam's own web module is in the closure, and no style is unresolved. Re-measured against9fbdfc592c: modules 4277 -> 4320, local 926 -> 970, minified 3,868,833 -> 3,812,418, a saving of 56,415 bytes. The two earlier readings and why they differ are kept in the census docstring. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb