Fourth Draft "Unsafe fp optimizations" project description.

[akbara]

>To give a warning on every instance of the use of a construct in the language,
>saying essentially "warning, this feature of the language should not be used"
good point.

> warning floating-point subtraction may cause loss of significance.
but imho, it's not as significant as an equality test.

doing something like if (blah_justcomputed == thevalue) could result into a lot of wasted time.

esp for the begginer who is learning how to implement various floating point intensive algorithms.
but then again, in most numerical anaylsis books; one of the first things they discuss are the numerical errors that come
up when writing software..

what do the others think about this warning?

-akbar A.


On Wed, 15 Aug 2001 18:22:00 -0400 (EDT)
dewar@gnat.com wrote:

> <<test.c:4: warning: comparing floating point with == or != is unsafe
> >>
> 
> This is an annoying warning, because it is nonsense. There are many
> situations in which it makes perfectly good sense to compare floating
> point numbers if you know what you are doing. Remember there is nothing
> "approximate" about floating-point, it is a well defined axiomatic system
> in which you can reason about your programs, and equality tests make
> perfect sense.
> 
> yes, they can be misused, but so can subtraction, I don't think you want
> to give warnings like:
> 
>   a = b - c;
> 
> warning floating-point subtraction may cause loss of significance.
> 
> It is particularly annoying that the messages says *is* unsafe, instead
> of something like "may be" unsafe.
> 
> Warnings should be about constructs in the language that are legal, but
> in some situations are likely to be clear programming errors.
> 
> To give a warning on every instance of the use of a construct in the language,
> saying essentially "warning, this feature of the language should not be used"
> 
> is contentious from the start. Why do you think floating-point equality
> was allowed in C (or in Fortran, Ada, COBOL, PL/1, Algol, and virtually
> every other programming language ever designed)? Presumably the writer
> of this warning thinks that was a mistake.
> 
> So it is almost never justified to give a warning for all uses of a feature
> in a language, it is even less justified when the warning comes from
> misguided ignorance :-)


-- 

"it's called Mario Twins, and I WANT TO PLAY IT"
http://vertexabuse.cjb.net