Re: [HACKERS] Re: Call for port testing on fmgr changes -- Results!

On Sun, 25 Jun 2000, Tom Lane wrote:

> > with this patch applied all regression tests pass just fine (save for
> > geometry of course with its standard off by one in nth decimal place
> > difference).
>
> Probably we should write that off as a platform issue and create an
> Alpha-specific expected-output file for geometry. See the documentation
> about platform-specific files, and please send along a patch to add one.

I remember finding a geometry.out file in the expected directory
that did match. Seems to me it was a Solaris/Sun one.... Anyway, I will
look into it and generate a patch to rid ourselves of that annoyance.
Probably will be next Sunday before it happens.
As for the comments about different geometry outputs with
different optimization levels, the linux_alpha template is set to use -O2,
which seems pretty standard for safe, yet useful optimization. Though I
have had reports that newer versions of gcc (from Mandrake 7.1) break
pgsql (spinlocks get stuck) with that optimization level. My alpha is
still running Debian 2.1 and so my gcc is a little old. Probably will
upgrade before the end of the summer and have to deal with that issue
then. Yuck. :( Of course, if some one wants to beat me to it, feel free :)

> > Additionally, I do not see how this patch could break other
> > platforms. At worst, it is a minor slow down that might even be optimized
> > out by some compiliers when they see that sizeof(AbsoluteTime) ==
> > sizeof(time_t). I will defer to the core developers on how you want to
> > apply this patch to the source tree (i.e. with #ifdef alpha && linux or as
> > above).
>
> No, we should just apply it as is, no #ifdef. There are going to be
> more and more platforms with 64-bit time_t.

No problem here with that, go ahead and apply it.

---------------------------------------------------------------------------
| "For to me to live is Christ, and to die is gain." |
| --- Philippians 1:21 (KJV) |
---------------------------------------------------------------------------
| Ryan Kirkpatrick | Boulder, Colorado | http://www.rkirkpat.net/ |
---------------------------------------------------------------------------