Re: [Gm2] gm2-20190709 on gcc-9.1.0 build failure

[Gaius Mulley]

john o goyo <address@hidden> writes:

> On 07/26/19 18:34, Gaius Mulley wrote (in part):
>> john o goyo <address@hidden> writes (in part):
>>> Indeed it has, Gaius, and back to the old ShiftLeft problem, which I
>>> will track one in due course.
>> Hi John,
>>
>> thanks for the reminder - I think adding extra regression tests to slice
>> up ShiftLeft would be interesting - hopefully will have a few written in
>> the next few days,
>
> Thank you, Gaius, that would be much appreciated.  The fact is, I
> quite ignorant of the inner workings of gcc.  Hunting down the problem
> would take me an enormous amount of time.

Hi John,

on my list of future changes would be to change the implementation
of large sets.  Currently they are a struct of BITSET fields - but this
would be better implemented as a pointer to array [0..maxelement] of
packed bool and callee copy for non var parameter - similar to how unbounded
arrays are implemented.  I suspect these changes would fix the sparc
shiftleft - just by the redesign alone.

But a number of tiny bitset regression tests around shift left would be
useful for both implementations,


regards,
Gaius