Both test scripts call rm without -f, so a fresh clone starts with an error

[dmccoystephenson]

run_tests.sh:2 runs rm ./tests_executable and run_tests_asan.sh:2 runs rm ./tests_executable_asan. Neither passes -f, and neither script sets set -e.

On a fresh clone the binary has never been built, so the first line of either script prints rm: cannot remove './tests_executable': No such file or directory before the build proceeds. Both binaries are listed in .gitignore, confirming they are never committed and are therefore always absent on a first checkout.

The consequence is cosmetic rather than fatal, since execution continues past the failed rm, but the noise is misleading to anyone running the suite for the first time and it leaves a nonzero status in the middle of what should be a clean run.

Suggested fix: pass -f to both rm invocations. Adding set -e to both scripts would be a reasonable companion change, so that a compile failure stops the script instead of falling through to run a binary that was just deleted, though that is a larger behavioral change and could be handled separately.

This issue body was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).

---

drafted by Claude on behalf of Daniel Stephenson

[dmccoystephenson]

This issue was reviewed during triage this cycle and intentionally deferred; the reasoning is recorded here so it does not need to be rediscovered on each pass.

The fix edits run_tests.sh and run_tests_asan.sh. In the automated environment these cycles run in, the compiled test binary cannot be executed — make is permitted, but ./tests_executable and a direct g++ invocation are both denied by the session's command allow-list. A change to the test tooling is therefore as unverifiable here as a change to src/: the corrected scripts could not be run to confirm that the rm -f change behaves as intended and that the rest of each script still functions. Such a change would have to be handed off unverified rather than merged.

Bundling it into this cycle's documentation change (#40, merged as #41) would additionally have blocked that verifiable change from merging.

The fix itself remains correct as described and is small; it is best applied where the suite can actually be executed once.

This comment was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).

---

drafted by Claude on behalf of Daniel Stephenson