Обсуждение: Solaris 8 regression test failure with 8.0.0beta5

Here are the diffs for the regression test failures on Solaris 8.
The tests work fine on Redhat9 and Redhat Enterprise Linux 3.

Ken Marshall

*** ./expected/errors.out    Sat Mar 13 22:25:17 2004
--- ./results/errors.out    Tue Nov 23 14:09:45 2004
***************
*** 297,303 **** -- Check that division-by-zero is properly caught. -- select 1/0;
! ERROR:  division by zero select 1::int8/0; ERROR:  division by zero select 1/0::int8;
--- 297,304 ---- -- Check that division-by-zero is properly caught. -- select 1/0;
! ERROR:  floating-point exception
! DETAIL:  An invalid floating-point operation was signaled. This probably means an out-of-range result or an invalid
operation,such as division by zero. select 1::int8/0; ERROR:  division by zero select 1/0::int8;
 

======================================================================

*** ./expected/transactions.out    Mon Sep 13 15:09:51 2004
--- ./results/transactions.out    Tue Nov 23 14:09:50 2004
***************
*** 290,296 **** BEGIN;   SAVEPOINT one;   SELECT 0/0;
! ERROR:  division by zero   SAVEPOINT two;    -- ignored till the end of ... ERROR:  current transaction is aborted,
commandsignored until end of transaction block   RELEASE SAVEPOINT one;      -- ignored till the end of ...
 
--- 290,297 ---- BEGIN;   SAVEPOINT one;   SELECT 0/0;
! ERROR:  floating-point exception
! DETAIL:  An invalid floating-point operation was signaled. This probably means an out-of-range result or an invalid
operation,such as division by zero.   SAVEPOINT two;    -- ignored till the end of ... ERROR:  current transaction is
aborted,commands ignored until end of transaction block   RELEASE SAVEPOINT one;      -- ignored till the end of ...
 
***************
*** 364,370 ****     DECLARE c CURSOR FOR SELECT unique2/0 FROM tenk1;     SAVEPOINT two;         FETCH 10 FROM c;
! ERROR:  division by zero     ROLLBACK TO SAVEPOINT two;     -- c is now dead to the world ...         FETCH 10 FROM
c;
--- 365,372 ----     DECLARE c CURSOR FOR SELECT unique2/0 FROM tenk1;     SAVEPOINT two;         FETCH 10 FROM c;
! ERROR:  floating-point exception
! DETAIL:  An invalid floating-point operation was signaled. This probably means an out-of-range result or an invalid
operation,such as division by zero.     ROLLBACK TO SAVEPOINT two;     -- c is now dead to the world ...         FETCH
10FROM c;
 

======================================================================

*** ./expected/plpgsql.out    Tue Nov 16 12:10:15 2004
--- ./results/plpgsql.out    Tue Nov 23 14:10:17 2004
***************
*** 1826,1837 ****  select trap_zero_divide(0); NOTICE:  should see this
! NOTICE:  caught division_by_zero
!  trap_zero_divide 
! ------------------
!                -1
! (1 row)
!  select trap_zero_divide(100000); NOTICE:  should see this NOTICE:  should see this only if 100000 <> 0
--- 1826,1834 ----  select trap_zero_divide(0); NOTICE:  should see this
! ERROR:  floating-point exception
! DETAIL:  An invalid floating-point operation was signaled. This probably means an out-of-range result or an invalid
operation,such as division by zero.
 
! CONTEXT:  PL/pgSQL function "trap_zero_divide" line 6 at assignment select trap_zero_divide(100000); NOTICE:  should
seethis NOTICE:  should see this only if 100000 <> 0
 

======================================================================

Kenneth Marshall <ktm@is.rice.edu> writes:
> Here are the diffs for the regression test failures on Solaris 8.
> The tests work fine on Redhat9 and Redhat Enterprise Linux 3.

... and most other platforms ...

>   select 1/0;
> ! ERROR:  floating-point exception
> ! DETAIL:  An invalid floating-point operation was signaled. This probably means an out-of-range result or an invalid
operation,such as division by zero.
 

This query would invoke the following code in int4div:
   if (arg2 == 0)       ereport(ERROR,               (errcode(ERRCODE_DIVISION_BY_ZERO),
errmsg("divisionby zero")));
 
   result = arg1 / arg2;

I suggest directing some strong words to your compiler vendor.  If the
"if" test doesn't complete before the subsequent division causes a trap,
that is not our problem.
        regards, tom lane

Tom Lane <tgl@sss.pgh.pa.us> writes:

> This query would invoke the following code in int4div:
> 
>     if (arg2 == 0)
>         ereport(ERROR,
>                 (errcode(ERRCODE_DIVISION_BY_ZERO),
>                  errmsg("division by zero")));
> 
>     result = arg1 / arg2;
> 
> I suggest directing some strong words to your compiler vendor.  If the
> "if" test doesn't complete before the subsequent division causes a trap,
> that is not our problem.

I have a vague memory that IEEE floats have zero coming in two flavours, 0 and
-0. I wonder if it's possible merely comparing against 0 isn't covering all
the possible cases that can trigger a division by zero trap.

-- 
greg

Greg Stark <gsstark@mit.edu> writes:
> Tom Lane <tgl@sss.pgh.pa.us> writes:

>> This query would invoke the following code in int4div:
>> 
>> if (arg2 == 0)

> I have a vague memory that IEEE floats have zero coming in two
> flavours, 0 and -0. I wonder if it's possible merely comparing against
> 0 isn't covering all the possible cases that can trigger a division by
> zero trap.

(a) This is an integer, not a float.

(b) The IEEE spec requires that 0 and -0 compare equal.
        regards, tom lane