Description
After #130739 I noticed that whenever I modify run-make
tests (on Windows), stage 1 cargo gets unnecessary rebuilt which also triggers rebuilds for other tools like stage 0 rustdoc. I'm pretty sure something in my implementation is wrong w.r.t. to properly caching, but I didn't invest time in figuring out why yet, hence opening this bug to track it.
The build sequence is something like stage 0 std -> stage 1 rustc -> stage 1 std -> stage 1 tool cargo -> stage 0 compiletest / run-make-support / rustdoc.
Example repro steps:
- Run
./x test run-make --stage 1
-> make sure stage 1 cargo is already built. - Modify a
tests/run-make
rmake.rs test. - Observe that stage 1 cargo gets rebuilt even though I didn't modify it.