Re: BUG #14722: Segfault in tuplesort_heap_siftup, 32 bit overflow

David Gould <daveg(at)sonic(dot)net> writes:
> Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> ... right. There haven't been any non-twos-complement machines in the
>> wild for probably 50 years, and even if there were, this would be *way*
>> down the list of problems you'd have to fix to get Postgres to run on
>> one of them.

> Not quite 50 years. In 1979 had the "pleasure" of working at Bechtel on a
> Univac 1110. Univac 1100 seris are ones-complement (with both positive
> and negative zero!) with 36 bit longs, 18 bit ints and depending on character
> mode either 9 bit ASCII or 6 bit FIELDDATA chars.

Fun stuff. Other than the ones-complement choice, this smells quite a bit
like the PDP-10 gear I used to use back when.

> Not even one year. UNISYS are still marketing this architecture as the UNISYS
> ClearPath IX series, you can order one today.

The recent models claim to be pure Intel though; if they're still
supporting the 1100 architecture, it must be through emulation.

> Still, I think it is safe to wait until someone actually pays for a
> postgresql port before considering ones-complement issues.

Yeah. I actually suspect that the weird word size would be a much bigger
headache for us than the ones-complement business. The other small
problem is that as far as I could find, they never went past a 24-bit
address space, which would make for at most 64MB worth of memory (with
9-bit "bytes"). In principle you could probably run modern Postgres
with so little RAM, but it wouldn't be of any real use.

regards, tom lane