Inconsistent tuplet-number-kneed-beam-horizontal-fit.ly results?

[Devon Schudy]

I'm getting unpredictable output from
input/regression/tuplet-number-kneed-beam-horizontal-fit.ly.

Sometimes the second tuplet number is in bracket position below the
staff, and sometimes it overlaps the beam. When I run this test via
make check, it may be in either position, but when I run it
individually (even via lilypond-book with the same options), it always
overlaps the beam.

In both cases the expected "not enough space for tuplet number against
beam" programming_error happens, so it's trying to follow the bracket
instead of the beam. (By the way, shouldn't this be a warning instead
of a programming_error?) Only the y position changes.

Does this happen for anyone else?

tuplet-number-kneed-beam-horizontal-fit.png
Description: PNG image

tuplet-number-kneed-beam-horizontal-fit.png
Description: PNG image