At 19:20 +0200 on 7/12/98, The Hermit Hacker wrote:
> Just checked the above...the files that are being reported as
> "missing" are those that are supposed to be created automagically when you
> do a 'make all'...see the input/* and output/* directories for the
> 'template' files for them.
>
> If you do a 'make clean all', does it give any errors when trying
> to create the files?
OK. It took me some time to synchronize my schedule to my sysadmin's. It
then took us some more time to realize that you were referring to the make
of the general postgres package and not the make of the regression test.
Once through that, your suggestion directed us to the problem - lack of
permissions in the regression directory, of all things.
To make a long story short, we succeeded in compiling and running postgres
and the regression tests on solaris 2.5.1. We needed flex and bison, but
eventually, not gcc. The Sun cc (not bundled) does a better job of
optimising, and gives the same regression results.
One thing still bugs me about the regression results - I can understand the
"fails" with float8, geometry, and abstime + its derivatives. But the int8
datatype seems to have a major failure. This type did not exist in 6.2.1,
my currently-effective version, so I have no comparison with past results.
The int8 test invariably displays '-268449328' for all values stored. For
example, the beginning of the result file in the regression test is:
QUERY: CREATE TABLE INT8_TBL(q1 int8, q2 int8);
QUERY: INSERT INTO INT8_TBL VALUES('123','456');
QUERY: INSERT INTO INT8_TBL VALUES('123','4567890123456789');
QUERY: INSERT INTO INT8_TBL VALUES('4567890123456789','123');
QUERY: INSERT INTO INT8_TBL VALUES('4567890123456789','4567890123456789');
QUERY: INSERT INTO INT8_TBL VALUES('4567890123456789','-4567890123456789');
QUERY: SELECT * FROM INT8_TBL;
q1| q2
- ----------+----------
- -268449328|-268449328
- -268449328|-268449328
- -268449328|-268449328
- -268449328|-268449328
- -268449328|-268449328
(5 rows)
It appears that the stored values are OK. That is, in the following test
(again taken from the results/int8.out file), one can see that the values,
once converted to float8, are OK.
QUERY: SELECT '' AS five, q1, float8(q1) FROM INT8_TBL;
five| q1|float8
- ----+----------+--------------------
|-268449328|123
|-268449328|123
|-268449328|4.56789012345679e+15
|-268449328|4.56789012345679e+15
|-268449328|4.56789012345679e+15
(5 rows)
So, there is some bug in the display procedure for int8, rather than an
internal problem. Is this true for all ports? Just for Solaris? Just for
Solaris 2.5.1? In any case, a fix is required. Maybe added to 6.4.1?
Herouth
- --
Herouth Maoz, Internet developer.
Open University of Israel - Telem project
http://telem.openu.ac.il/~herutma
