cwd target doesn't work with ibazel

[alexeagle]

A customer pointed out that the pattern documented on the README
https://github.com/theoremlp/rules_multitool?tab=readme-ov-file#running-tools-in-the-current-working-directory
doesn't work correctly with some programs like ibazel (bazel-watcher) because zombie processes get left behind.
We think this post from @lberki is related https://groups.google.com/g/bazel-discuss/c/I6TWUv8AI5o
I think this can be reproduced by running some server target under ./tools/ibazel //my/server:target, then when you ctrl-c (SIGINT) you'll still see the server is running.

So we changed our aspect init aspect-build/aspect-workflows-template#87
to still stamp out a script that DOES function correctly:

#!/bin/sh
target="@multitool//tools/$(basename "$0")"
bazel 2>/dev/null build "$target" && exec $(bazel info execution_root)/$(bazel 2>/dev/null cquery --output=files "$target") "$@"

Personally I think bazel run is still just busted in a way rules_multitool shouldn't try to fix.

[mark-thm]

Ah, that's a bummer. I wasn't able to find a current issue on Bazel's GitHub issues tracker, I'll file one.

[mark-thm]

I tried to build a minimal reproduction of bazel run propagating SIGINT but was able to successfully trap SIGINT using the example from the linked mailing list. It seems like bazel run (at least in Bazel 7.0.0+) is propagating SIGINT.

I do see that ibazel is leaving dangling processes when run through multitool's cwd target, but it's less clear to me that bazel run is broken?

[mark-thm]

When running your provided script I see the same dangling bazel processes -- so it seems like this is really an ibazel bug.

[alexeagle]

Interesting, @walkerburgin confirmed they saw a problem but I don't know what Bazel version. Do you have time to make a repro?

[mark-thm]

Running latest ibazel directly (no bazel, shell, or multitool involved) also leaves dangling processes on a ctrl+c.

[alexeagle]

Yeah it's an ibazel bug, it so happens that calling it through bazel run exposes it while invoking the ibazel binary directly does not. So I think the only change here should be a warning in that README not to use that pattern with ibazel.