Thread: numeric test on RiscPC

numeric test on RiscPC

From
Patrick Welche
Date:
I have just found to my surprise:

============== running regression test queries        ==============
parallel group (13 tests):  char name int2 text float4 oid int4 varchar int8 float8 boolean bit numeric

  PID USERNAME PRI NICE   SIZE   RES STATE      TIME   WCPU    CPU COMMAND
27747 prlw1     72    4   596K  276K RUN      277.7H 98.10% 98.10% sh

277.7H being just over 11.5 days!

Have any of you tried

 PostgreSQL 7.2 on acorn32-unknown-netbsd1.5ZB, compiled by GCC egcs-1.1.2

on an Acorn RiscPC with a SA-110? (This is the well known "halting problem" :) )

Cheers,

Patrick

Re: numeric test on RiscPC

From
Tom Lane
Date:
Patrick Welche <prlw1@newn.cam.ac.uk> writes:
> I have just found to my surprise:
> ============== running regression test queries        ==============
> parallel group (13 tests):  char name int2 text float4 oid int4 varchar int8 float8 boolean bit numeric

>   PID USERNAME PRI NICE   SIZE   RES STATE      TIME   WCPU    CPU COMMAND
> 27747 prlw1     72    4   596K  276K RUN      277.7H 98.10% 98.10% sh

> 277.7H being just over 11.5 days!

> Have any of you tried
>  PostgreSQL 7.2 on acorn32-unknown-netbsd1.5ZB, compiled by GCC egcs-1.1.2
> on an Acorn RiscPC with a SA-110? (This is the well known "halting problem" :) )

Looks like Acorn's shell has the same bug documented to exist in HPUX's
shell (see doc/FAQ_HPUX :-() ... it gets confused when it has to manage
more than about a dozen child processes.

On HPUX I can work around this by telling pg_regress to use ksh instead.
If you have any other shells besides plain sh, give them a try.

            regards, tom lane

Re: numeric test on RiscPC

From
Patrick Welche
Date:
On Mon, Apr 08, 2002 at 10:18:23AM -0400, Tom Lane wrote:
> Patrick Welche <prlw1@newn.cam.ac.uk> writes:
...
> > Have any of you tried
> >  PostgreSQL 7.2 on acorn32-unknown-netbsd1.5ZB, compiled by GCC egcs-1.1.2
> > on an Acorn RiscPC with a SA-110? (This is the well known "halting problem" :) )
>
> Looks like Acorn's shell has the same bug documented to exist in HPUX's
> shell (see doc/FAQ_HPUX :-() ... it gets confused when it has to manage
> more than about a dozen child processes.
>
> On HPUX I can work around this by telling pg_regress to use ksh instead.
> If you have any other shells besides plain sh, give them a try.

Thanks for the tip! Even have the timezone errors mentioned in there. Now
it gets as far as create_function_2. Hunting will be interesting: the only
architecture dependent thing I can see in sh is:

.if (${MACHINE_ARCH} == "powerpc") || \
    (${MACHINE_CPU} == "arm")
TARGET_CHARFLAG= -DTARGET_CHAR="u_int8_t"
.else
TARGET_CHARFLAG= -DTARGET_CHAR="int8_t"
.endif

in the makefile.. (acorn32==arm) What is the HPUX fix?

Cheers,

Patrick

Re: numeric test on RiscPC

From
Tom Lane
Date:
Patrick Welche <prlw1@newn.cam.ac.uk> writes:
> On Mon, Apr 08, 2002 at 10:18:23AM -0400, Tom Lane wrote:
>> Looks like Acorn's shell has the same bug documented to exist in HPUX's
>> shell (see doc/FAQ_HPUX :-() ... it gets confused when it has to manage
>> more than about a dozen child processes.

> it gets as far as create_function_2. Hunting will be interesting: the only
> architecture dependent thing I can see in sh is:

I doubt it's architectural at all ... try looking for fixed-size
arrays...

            regards, tom lane