Re: whack-a-regression

[David Kastrup]

Keith OHara <address@hidden> writes:

> Keith OHara <k-ohara5a5a <at> oco.net> writes:
>
>> You could experiment to understand better the situations that lead to the
>> out-of-memory crash under Windows --- issue 3432 and other reports on bug-
>> lilypond yesterday.  This makes version 2.17.21 useless on Windows.
>
>> Looking at the commits since version 2.17.20, the only one that seems 
>> related is my change to outside-staff-padding, but I cannot imagine a way
>
> That one was the patch for issue 2910.
> Maybe it would help to experiment to find whether 'outside-staff-padding 
> and 'horizontal-skyline-padding affect the problem.
>
> There is also 
>  Issue3383: old-straight-flag + smaller Stem.thickness gives no output and
> that rewrites a lot of skylining code, but the new code looked good.

It's worth noting that one iteration of the issue 3383 patch had the
"memory grows indefinitely" syndrome, and the main difference to the
not-indefinitely memory gobbling next iteration was exchanging one <=
for a < regarding floating point operations.  While I'd agree with "the
new code looked good", actually better than the old code, this kind of
not really satisfactorily explained sensitivity looks creepy enough.

I'm not sure the same sensitivity remained with the last iteration, but
it's worth remembering.

-- 
David Kastrup