Experimental Physics and
Industrial Control System

Dirk Zimoch <[email protected]> · Wed, 06 Jun 2012 12:09:20 +0200

Eric Norum wrote:

On Jun 5, 2012, at 9:46 AM, Andrew Johnson wrote:

Hi Eric,

On Jun 5, 2012, at 6:11 AM, Dirk Zimoch wrote:

I think adding something like this to epicsStrtod should fix the problem:

if (epicsStrnCaseCmp("0x", cp, 2) == 0) {
return (double)strtoul(str, endp, 16);
}

I'll be using strtol() instead in the 3.15 code, so -0x1 works.

Right.

I had assumed that the primary application for hex numbers would be bit

masks. Signed masks do not make much sense but 0x80000000 does make

sense. Maybe we should test for minus and then either call strtol() or

strtoul().

On 2012-06-05 Eric Norum wrote:

The code with that change doesn't completely match the semantics of strtod
as presented in the C99 standard, but it's certainly better than the
current code.

What else are we missing?  I don't have the C99 standard to look at here.

A  '0x' number can have a fractional part and an exponent.  This seems pretty esoteric so its absence is, IMHO, minor.

Nobody right in his mind uses hexadecimal floating point numbers. ;-)

BTW -- does vxWorks still not supply a working strtod?  I
think that it is the only system that doesn't just define epicsStrtod as
strtod.

The wording in the vxWorks 6.8 manpage is exactly the same as Dirk posted from

vxWorks 5.5.

:-(

My argument is: Yes, vxWorks 5.5 still does not supply a working strtod.

Changing the vxWorks version is not as simple as upgrading Ubuntu.

No, we use our version of epicsStrtod() on Windows, Cygwin, vxWorks and

Solaris. I have a branch for 3.15 where I've been working on the numeric

conversions, I'll add some tests for the native strtod().

Yuck.
After over 12 years they still can't produce a strtod that matches the standard? (!!!)

They are WindRiver (or rather Intel now). They make their own standard.

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System