Ar an cúigiú lá déag de mí Méan Fómhair, scríobh Mats Lidell:
> Aidan Kehoe writes:
> I wonder is it just as simple as the redisplay structures being
> inconsistently marked as GC roots through the kkcc descriptions? That might
> fit with my never seeing the crash on this TTY-only NEWGC build. Though it
> should provoke the same crashes on a KKCC non-NEWGC build, which we don’t
Are you looking at this problem? From your description it sound like an easy
thing to test but that is just me counting the number of lines describing the
issue! (One line describing the problem = easy thing to fix!) Or is there not
an inconsistency that should be viewed as bug?
I mean that the reason we don't see the crashes might be that it is hard to
get this error in the first place. I can't recreate it on the machines that I
have available so it is tricky to get and of course even trickier to know if
it is solved.
Maybe, if I could get a patch with this fix I could get the reported, who gets
the error, to test this!?
I have been looking at the problem. I have a reasonable working hypothesis
for it, which is that cmotion_display_lines in redisplay.c should be marked
for garbage collection but isn’t, and that with NEWGC the shorter intervals
between GC mark activity make the garbage more likely to be encountered.
I have the bulk of a patch written, but there are six relevant combinations
of configure options it has to work with (NEWGC + KKCC + PDUMP, oldgc + KKCC
+ PDUMP, oldgc + KKCC + olddump, oldgc + oldmark + PDUMP, oldgc + oldmark +
olddump, NEWGC + KKCC + olddump) and so I haven’t yet got it to the point
where I can commit it.
I also have an exam at the end of the month so have been studying for that!
‘As I sat looking up at the Guinness ad, I could never figure out /
How your man stayed up on the surfboard after forty pints of stout’