Prev: Oversight in CREATE FUNCTION + EXPLAIN?
Next: plruby code and postgres ?
From: Boszormenyi Zoltan on 19 Nov 2009 14:12
Boszormenyi Zoltan �rta:
> Hi,
>
> my question is that what platform were these
> functions developed and tested?
>
> We have come across a value that fails a NOT NULL
> constraint upon INSERT under HP-UX/IA64, but not
> under x86-64 Linux. The value in question is
> 1.9999999999999998 assigned to a "double" variable.
> Under HP-UX/IA64, testing with risnull() from
> the application indeed returns true, but under
> Linux/x86-64 returns false.
>
> I will test rsetnull() results on real Informix under
> HP-UX/IA64.
>

I have tested it under ESQL/C on HP-UX/ia64 and
this happened:
- rsetnull() on a double value creates
FF FF FF FF FF FF FF FF
- the value causing the error above is
3F FF FF FF FF FF FF FF

It seems that this function in ecpglib/misc.c has
an off-by-one bug as it's interpreted by the HP-UX CC:

static bool
_check(unsigned char *ptr, int length)
{
for (; length > 0 && ptr[--length] == 0xff;);
if (length <= 0)
return true;
return false;
}

I suspect that GCC does the "--length" after checking
"length > 0" and before checking the "ptr[...] == 0xff",
but HP CC does it before checking "length > 0".

The attached patch solves the problem.

Best regards,
Zolt�n B�sz�rm�nyi

--
Bible has answers for everything. Proof:
"But let your communication be, Yea, yea; Nay, nay: for whatsoever is more
than these cometh of evil." (Matthew 5:37) - basics of digital technology.
"May your kingdom come" - superficial description of plate tectonics

----------------------------------
Zolt�n B�sz�rm�nyi
Cybertec Sch�nig & Sch�nig GmbH
http://www.postgresql.at/

From: Tom Lane on 19 Nov 2009 14:33
Boszormenyi Zoltan <zb(a)cybertec.at> writes:
> for (; length > 0 && ptr[--length] == 0xff;);

> I suspect that GCC does the "--length" after checking
> "length > 0" and before checking the "ptr[...] == 0xff",
> but HP CC does it before checking "length > 0".

If it does, that is *unquestionably* a bug in HP's CC and should be
reported to them. However, the code is sufficiently unreadable to
be worth rewriting anyhow. Your suggestion is an improvement but
personally I'd plump for

int i;

for (i = 0; i < length; i++)
if (ptr[i] != 0xff)
return false;
return true;

regards, tom lane

--
Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

From: Andrew Dunstan on 19 Nov 2009 14:46


Tom Lane wrote:
> Boszormenyi Zoltan <zb(a)cybertec.at> writes:
>
>> for (; length > 0 && ptr[--length] == 0xff;);
>>
>
>
>> I suspect that GCC does the "--length" after checking
>> "length > 0" and before checking the "ptr[...] == 0xff",
>> but HP CC does it before checking "length > 0".
>>
>
> If it does, that is *unquestionably* a bug in HP's CC and should be
> reported to them.
>

Wow, I recall fighting HP over a bad compiler bug (although not as bad
as this would be) 15 years ago. Their official response amounted to "we
don't care and we're not going to fix it". Maybe not much has changed.

cheers

andrew

--
Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

From: Boszormenyi Zoltan on 19 Nov 2009 15:32
Tom Lane �rta:
> Boszormenyi Zoltan <zb(a)cybertec.at> writes:
>
>> for (; length > 0 && ptr[--length] == 0xff;);
>>
>
>
>> I suspect that GCC does the "--length" after checking
>> "length > 0" and before checking the "ptr[...] == 0xff",
>> but HP CC does it before checking "length > 0".
>>
>
> If it does, that is *unquestionably* a bug in HP's CC and should be
> reported to them.

Is it *really* a bug? I recalled a comment from my C teacher
in '92 or '93 about this exact issue, that the prefix/postfix
increment/decrement operators are executed in the
statement in an implementation-defined order, i.e. they
can be freely reordered or placed anywhere in the
expression, provided that the postfix operator's evaluation
is earlier than the usage of the variable it's used on and
evaluation is later than the variable usage in the postfix case.
This means that their usage has to be minimized so the
result is unambiguous. I.e. in the common usage:

str1[pos1++] = str2[pos2++];

these execution orders are possible and all give the same result:

1. evaluate str2[pos2]
increment pos2
assign the above value to str1[pos1]
increment pos1
or
2. evaluate str2[pos2]
assign the above value to str1[pos1]
increment pos2
increment pos1
or
3. evaluate str2[pos2]
assign the above value to str1[pos1]
increment pos1
increment pos2

In the case of
for (; length > 0 && ptr[--length] == 0xff;);
the different evaluation orders may give different
expression results.

But 17 years is a long time, the C language specification
has changed a lot. GCC definitely does the most sensible
order but I didn't know this behaviour is specified in the
C language.

> However, the code is sufficiently unreadable to
> be worth rewriting anyhow. Your suggestion is an improvement but
> personally I'd plump for
>
> int i;
>
> for (i = 0; i < length; i++)
> if (ptr[i] != 0xff)
> return false;
> return true;
>

Yes, it's better than my version.

Best regards,
Zolt�n B�sz�rm�nyi

> regards, tom lane
>
>


--
Bible has answers for everything. Proof:
"But let your communication be, Yea, yea; Nay, nay: for whatsoever is more
than these cometh of evil." (Matthew 5:37) - basics of digital technology.
"May your kingdom come" - superficial description of plate tectonics

----------------------------------
Zolt�n B�sz�rm�nyi
Cybertec Sch�nig & Sch�nig GmbH
http://www.postgresql.at/


--
Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

From: Boszormenyi Zoltan on 19 Nov 2009 15:36
Boszormenyi Zoltan �rta:
> Tom Lane �rta:
>
>> Boszormenyi Zoltan <zb(a)cybertec.at> writes:
>>
>>
>>> for (; length > 0 && ptr[--length] == 0xff;);
>>>
>>>
>>
>>
>>> I suspect that GCC does the "--length" after checking
>>> "length > 0" and before checking the "ptr[...] == 0xff",
>>> but HP CC does it before checking "length > 0".
>>>
>>>
>> If it does, that is *unquestionably* a bug in HP's CC and should be
>> reported to them.
>>
>
> Is it *really* a bug? I recalled a comment from my C teacher
> in '92 or '93 about this exact issue, that the prefix/postfix
> increment/decrement operators are executed in the
> statement in an implementation-defined order, i.e. they
> can be freely reordered or placed anywhere in the
> expression, provided that the postfix operator's evaluation
>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Above is the prefix case obviously...

> is earlier than the usage of the variable it's used on and
> evaluation is later than the variable usage in the postfix case.
> This means that their usage has to be minimized so the
> result is unambiguous.

--
Bible has answers for everything. Proof:
"But let your communication be, Yea, yea; Nay, nay: for whatsoever is more
than these cometh of evil." (Matthew 5:37) - basics of digital technology.
"May your kingdom come" - superficial description of plate tectonics

----------------------------------
Zolt�n B�sz�rm�nyi
Cybertec Sch�nig & Sch�nig GmbH
http://www.postgresql.at/


--
Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

 |  Next  |  Last
Pages: 1 2
Prev: Oversight in CREATE FUNCTION + EXPLAIN?
Next: plruby code and postgres ?