Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
de1f19e to
ffdd82e
Compare
|
This PR is blocked by mlc-ai/relax#343 and needs #192 to be merged |
|
This is amazing, finally will be able to run debug sessions on simulator! |
JiwaniZakir
left a comment
There was a problem hiding this comment.
The python vs python3 inconsistency in build-runtime.sh is a subtle bug: the availability check on line ~251 was changed to use python -m mlc_llm, but both mlc_llm package invocations inside the new iOS loop still call python3. This means the guard can pass on systems where only python3 exists (with python aliased elsewhere), yet the actual build commands will fail.
The in-place sed -i '' mutation of $MLC_LLM_SOURCE_DIR/ios/prepare_libs.sh is fragile — if the build is interrupted mid-loop, the file is left in a modified state with is_simulator set to whichever value was last written, making the next run produce incorrect output silently. If MLC_LLM_SOURCE_DIR is a git submodule, this also introduces dirty state. A safer pattern would be to copy the script to a temp location, apply the substitution there, and execute the copy.
Removing the CMake cache conflict detection block without replacing it with per-platform build directories is a regression risk. The BUILD_DIR is now $SCRIPT_DIR/build (shared across both simulator and device iterations), and the loop already handles cleaning it between runs — but for the Android path, the old guard that detected cross-platform cache contamination is now gone entirely.
This PR:
build-runtime.shscript to build for both iPhone and iOS simulator, outputs those tolib_iphoneandlib_iphonesim