* Skip redundant CI rebuilds for soulamz and OCR/WARP tooling. Refresh soulamz.inputs so prebuilts match current inputs, track OCR/WARP dex in git with tooling.inputs, build soulamz ABIs in parallel, and drop Actions tooling cache in favor of checkout prebuilts. Co-authored-by: Cursor <cursoragent@cursor.com> * Make soulamz and tooling input stamps path-independent. sha256sum was hashing absolute file paths, so CI and local machines computed different values and always rebuilt prebuilts. Co-authored-by: Cursor <cursoragent@cursor.com> * Store OCR/WARP dex prebuilts under prebuilts/ for reliable CI checkout. Gitignored tools/ and bundled-ocr/ paths were not always restored on CI runners. Relocate dex artifacts to a normal tracked directory and keep using path-independent input stamps. Co-authored-by: Cursor <cursoragent@cursor.com> * Use .dat extension for tracked OCR/WARP prebuilts. Global *.dex gitignore was preventing reliable checkout on CI runners; store dex bytes as .dat files instead. Co-authored-by: Cursor <cursoragent@cursor.com> * Log why OCR/WARP tooling prebuilts are rebuilt on CI. Co-authored-by: Cursor <cursoragent@cursor.com> * Hash prebuilt inputs from file contents only. Filename-bearing sha256sum output still diverged between CI and local machines for tooling sources; aggregate raw bytes instead. Co-authored-by: Cursor <cursoragent@cursor.com> * Debug tooling stamp at start of build step Co-authored-by: Cursor <cursoragent@cursor.com> * Hash prebuilt inputs with git hash-object for CI parity. Working-tree byte hashes differed on GitHub runners; git blob hashes match the committed tree on every checkout. Co-authored-by: Cursor <cursoragent@cursor.com> * Hash prebuilt inputs from git tree blob ids. Use ls-tree blob hashes so checkout line-ending filters cannot skew the stamp on CI runners. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com>
1 line
65 B
Text
1 line
65 B
Text
77c94253285326b89169abe9d55185ff17ad33921b34d55afb2ffee1a2a78507
|