One more follow-up on this. As we saw from the bug reports for the 1.16i pretest (67894, 67918), the random parallelization failures were back after this, when autom4te did not support subsecond-mtimes.
As far as I could tell, what was needed was to add am_cv_sleep_fractional_seconds=false which is what the previous sanity.m4 code did when a non-subsecond autom4te was detected. With this, my make (dist)check then started reliably succeeding again. (Though vastly more slowly, of course.) --- a/t/ax/test-defs.in +++ b/t/ax/test-defs.in @@ -193,6 +193,10 @@ case $MTIME_RESOLUTION in : else MTIME_RESOLUTION=1 + # we must not sleep for fractional seconds when autom4te does not + # support subsecond-mtimes, even when sleep supports it. See bug#67670. + am_cv_sleep_fractional_seconds=false + export am_cv_sleep_fractional_seconds fi ;; esac My actual commit has a lot of related verbiage, but that additional assignment is the only substantive change. Nothing to do here (comments welcome of course), but just wanted to get this into the bug archive ... --thanks, karl.