Robert Haas <robertmhaas@gmail.com> writes:
> On Fri, Mar 14, 2014 at 4:43 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> How is that leading to a crash? Well, this machine is 32-bit, so MAXALIGN
>> is only 4. This means it is possible for an odd-length message cum
>> message length word to not exactly divide the size of the shared memory
>> ring buffer, resulting in cases where an 8-byte message length word is
>> wrapped around the end of the buffer.
> Argh. I think I forced the size of the buffer to be MAXALIGN'd, but
> what it really needs is to be a multiple of the size of uint64.
After sleeping on it, I think what you're proposing here is to double down
on a wrong design decision. ISTM you should change the message length
words to be size_t (or possibly ssize_t, if you're depending on signed
arithmetic), which would let you keep using MAXALIGN as the alignment
macro. There is absolutely no benefit, either for performance or code
readability, in forcing 32-bit machines to use 64-bit message length
words. Indeed, by not using the same alignment macros as everywhere else
and not being able to use %zu for debug printouts, I think the only real
effect you're producing is to make the DSM/MQ stuff more and more randomly
unlike the rest of Postgres. Please reconsider while it's still not too
late to change those APIs.
regards, tom lane