gnuplot-beta

Re: Gentlemen, start your engines...

[Daniel J S. <dan...@ie...>]

Ethan Merritt wrote:
> On Friday 06 October 2006 04:36 pm, Daniel J Sebald wrote:
> 
>>Ethan Merritt wrote:
>>
>>>Digital Unix 4.0D (DECC compiler)
>>> - Builds OK, but ...
>>> - triggers floating exception in surface1.dem
>>>	(Bug #1572268)
>>> - triggers floating exception in image.dem when trying to
>>>   read deliberately wrong-endian binary data
>>>	(maybe we should remove this demo)
>>
>>What happens on the DEC with
>>
>>set samples 101
>>plot 1/x
> 
> 
> Nothing special.  The issue is not divide-by-zero or underflow.
> The wrong-endian exception is from trying to do anything at all
> with a  bit pattern that is not a valid floating point number.

I'm just wondering how generally the DEC behaves on an FPE.  If it happens that the FPE occurs inside

	return df_readbinary(v, max);

at the point of constructing the floating point value, we may be able to trap that and map the status to "undefined".

Dan

Re: Gentlemen, start your engines...

[<br...@ph...>]

Ethan Merritt wrote:

> What is the correct way to have ./configure figure this out?

There are two possibilities:

1) one that goes very much against the grain of what autoconf was 
designed for: a switch/case on (part of) the canonical platform name, as 
figured out by config.guess.

2) the fully autoconf-ish approach: check whether -ieee improves the 
behaviour of a test program, by compiling and running it twice.

3) the middle way: compile a test souce with an #ifdef __ALPHA__  or so, 
and if it's there, add -ieee to the CFLAGS.

For an examples of technique 3), see the existing macros in m4/*.m4.