Обсуждение: Patch: make behavior of all versions of the "isinf" function be similar

Hello, hackers!

While I was searching for a function which checks doubles for
infinity, I discovered a function "isinf" in a file src/port/isinf.c
where one of three versions returns different value for "-inf" ("1"
instead of "-1") comparing to the other two.

It seems concrete values (not just "if isinf(...)") are checked only
in float.c in float4out and float8out, but I am going to check for
concrete values in my another patch.

For systems with HAVE_FPCLASS the function returns the same result for
both "+inf" and "-inf". I can't test it on my WS, but I found only one
man page[1] where system header file "ieeefp.h" must be included for
ability to use "fpclass" function.

I guess nothing will be broken if that version of the function returns
the same results for input values as the other two.

Proposed patch makes that behavior.

P.S.: Should the patch be added to the next CF?

[1]https://docs.oracle.com/cd/E36784_01/html/E36874/fpclass-3c.html
--
Best regards,
Vitaly Burovoy

On Mon, Feb 1, 2016 at 8:13 AM, Vitaly Burovoy <vitaly.burovoy@gmail.com> wrote:
> While I was searching for a function which checks doubles for
> infinity, I discovered a function "isinf" in a file src/port/isinf.c
> where one of three versions returns different value for "-inf" ("1"
> instead of "-1") comparing to the other two.
>
> For systems with HAVE_FPCLASS the function returns the same result for
> both "+inf" and "-inf". I can't test it on my WS, but I found only one
> man page[1] where system header file "ieeefp.h" must be included for
> ability to use "fpclass" function.

That's the case on OSX, but on Linux -1 is returned for -Inf, and 1 for +Inf.

> I guess nothing will be broken if that version of the function returns
> the same results for input values as the other two.
> Proposed patch makes that behavior.

Actually, is there actually a reason to keep this file in the code
tree? Are there platforms that do not have isinf()? Even for Windows
environments using MSVC < 1800 this is emulated using _fpclass.
-- 
Michael

On Mon, Feb 1, 2016 at 2:38 PM, Michael Paquier
<michael.paquier@gmail.com> wrote:
> Actually, is there actually a reason to keep this file in the code
> tree? Are there platforms that do not have isinf()? Even for Windows
> environments using MSVC < 1800 this is emulated using _fpclass.

Looking at what is in the buildfarm, I noticed that isinf is provided
everywhere. Attached is a patch. Thoughts?
--
Michael

Michael Paquier <michael.paquier@gmail.com> writes:
> On Mon, Feb 1, 2016 at 2:38 PM, Michael Paquier
> <michael.paquier@gmail.com> wrote:
>> Actually, is there actually a reason to keep this file in the code
>> tree? Are there platforms that do not have isinf()? Even for Windows
>> environments using MSVC < 1800 this is emulated using _fpclass.

> Looking at what is in the buildfarm, I noticed that isinf is provided
> everywhere. Attached is a patch. Thoughts?

Two comments:

1. I don't think the buildfarm is sufficient evidence to conclude that
isinf.c is required nowhere.  It was in use as late as 2004, judging
by the git history, and I don't know of good reason to assume we do not
need it now.

2. POSIX:2008 only requires that "The isinf() macro shall return a
non-zero value if and only if its argument has an infinite value."
Therefore, any assumption about the sign of the result is wrong anyway,
and any patch that depends on it will be rejected, regardless of what
isinf.c does.  Or else, if you can convince us that such an assumption
is really valuable, we'd need isinf.c more not less so that we can
replace isinf() on platforms where it doesn't meet the stronger spec.
        regards, tom lane

On 1/31/16, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> 2. POSIX:2008 only requires that "The isinf() macro shall return a
> non-zero value if and only if its argument has an infinite value."
> Therefore, any assumption about the sign of the result is wrong anyway,
> and any patch that depends on it will be rejected, regardless of what
> isinf.c does.  Or else, if you can convince us that such an assumption
> is really valuable, we'd need isinf.c more not less so that we can
> replace isinf() on platforms where it doesn't meet the stronger spec.
>
>             regards, tom lane

Ok, then I'll use "is_infinite" from "float.c".
But why functions' (in "src/port/isinf.c") behavior are different? It
is a bit confusing…
--
Best regards,
Vitaly Burovoy

Vitaly Burovoy <vitaly.burovoy@gmail.com> writes:
> On 1/31/16, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> 2. POSIX:2008 only requires that "The isinf() macro shall return a
>> non-zero value if and only if its argument has an infinite value."

> Ok, then I'll use "is_infinite" from "float.c".

Yeah, that's good.

> But why functions' (in "src/port/isinf.c") behavior are different? It
> is a bit confusing…

Probably the authors of those different implementations were making it
match the behavior of whatever isinf() they had locally.  I don't feel
a need to change isinf.c --- it should be zero-maintenance at this
point, especially if we suspect it is no longer used anywhere.
        regards, tom lane