Auto merge of #129509 - ChrisDenton:hack-hack-hack, r=Mark-Simulacrum

Add a hack to workaround MSVC CI issues

I'm still attempting to figure out the issue but the failure rate atm seems really high so I'd like to propose an interim hack.

This adds a single delay of 5 minutes, which is absurdly long in CPU time but practically nothing compared to overall CI time.

src/ci/docker/host-x86_64/x86_64-gnu-tools/checktools.sh

@@ -55,6 +55,17 @@ case $HOST_TARGET in
     python3 "$X_PY" test --stage 2 src/tools/miri --target s390x-unknown-linux-gnu --test-args pass
     ;;
   x86_64-pc-windows-msvc)
+    # FIXME: This will rebuild miri because it's being built without the miri flags above.
+    # However, for some reason rebuilding is very likely to fail at the final stage when
+    # Cargo overwrites the existing hardlinks.
+    # So we introduce a pause here in the hope it fixes itself and display a list of process for
+    # diagnotistic purposes.
+    ps -W || true
+    echo "HACK: sleeping for 5 minutes"
+    sleep 5m
+    # Show the process list again in case the changes are interesting.
+    ps -W || true
+
     # Strangely, Linux targets do not work here. cargo always says
     # "error: cannot produce cdylib for ... as the target ... does not support these crate types".
     # Only run "pass" tests, which is quite a bit faster.