Thomas Lockhart <lockhart@alumni.caltech.edu> writes:
> Is the original issue support for 0x10... as the smallest integer, as
> opposed to -MAX_INT? As long as we continue to map the "reserved values"
> to the upper and lower range of allowed values so they are unlikely to
> appear under normal circumstances, the change should be OK.
I think that the original problem was that Andreas was seeing a compiler
codegen bug on AIX, having to do with the comparisonfoo > INT_MIN
generated by the AbsoluteTimeIsReal macro. I think he was seeing that
the compiler insisted on generating an unsigned compare, explicit casts
to signed datatypes notwithstanding :-(.
The proposed fix was to recode the macro's test as foo != INT_MIN,
thereby avoiding the issue of whether the comparison is signed or not.
To do that, we needed to make NOSTART_ABSTIME be defined as INT_MIN
on all platforms, not only AIX. That seemed like a good general-purpose
approach to me anyway, since the intended meaning of 0x80000000 was very
unclear otherwise.
regards, tom lane
PS: I'm quite sure that I'd explicitly cc'd you on the prior discussion.
If you didn't see it, then you've lost personal mail, not only pghackers
traffic...