C execution breakage

@arghdos !108 (merged) (the C execution target) has provoked a bunch of breakage in the presence of caching. I'm sorry that the CIs didn't catch that; all the test runs had caching turned off. I've since added one run with caching enabled, so that we can tell apart cache-related from non-cache related failures. (see 975fd86b) See here for the fallout:

Job #39072 failed for 0b81242f

c5e07552, a commit immediately before that merge, does not exhibit these issues.