Re: [Gnu-arch-users] Nit

[Dustin Sallings]

On Wednesday, Oct 22, 2003, at 05:35 US/Pacific, Andrew Suffield wrote:

> > Since perl has no number types,
>
> Totally wrong. Fundamentally so, even. Perl has three primitive
> numeric types (two integer, one floating) plus a bunch of interesting
> things done using magic, several of which are distributed in the core.

	This is clearly a matter of perspective.  The closest type perl has is 
a scalar which is sometimes interpreted as a number if the conditions 
are correct.  In my example, "0" by itself is considered a number for 
the condition, but "0.0" is not.  Math is possible on both of them.  
String manipulation is possible on both of them.  In fact, the only way 
to tell if a scalar is a number is to perform string pattern matching 
operations on it.  Languages with numeric types don't have this issue.

> Only if you're a lunatic. These two statements:
>
> "Replace functions with equivalents which succeed or die"
>
> "Replace functions with equivalents which die on error"
>
> are obviously not the same thing.

        Uh, yeah, obviously.

> I've read the Fatal documentation carefully, and it most definitely
> describes what it does accurately and precisely, and not what you say.

	I've read the documentation carefully as well, along with the source.  
However, such modules shouldn't be so obscure that this is necessary.  
It gives open and close as examples.  It should be safe to assume that 
read and syswrite can be used the same way.  They cannot.

> >         Surely the authoritative online documentation should be considered
> > trustworthy.  It makes no suggestion that it is not the appropriate
> > tool for the job.  It seems to do quite the opposite, to me.
>
> It makes no suggestion about appropriateness at all, and it almost
> certainly shouldn't.

	I agree, documentation *definitely* shouldn't give you a clue as to 
when you should actually do something, or even that it's a bad idea.  
It should only tell you that the particular thing being documented is a 
good example of how to do something and leave it to intuition to know 
that it's not good in practice.

	Anyway, this whole perl topic has distracted greatly from the original 
point.

	You said that only a moron would write code that doesn't check for all 
errors...er, unless it's writing to stdout or stderr since ENOSPC is 
obviously a temporary error and applications should queue up writes and 
try again later.  Except yours since you specifically avoid checking 
the results in the first place.

	Luckily, most of the languages I use will help with this situation.  
Python, ocaml, eiffel, java, etc... will quickly inform me with an 
exception when I attempt to perform a write and that write fails, 
regardless of file descriptor.  Tom has provided a similar concept in 
hackerlab so write errors won't be dropped there, either.

	Anyway, this isn't going anywhere.  You won't convince me that it's OK 
to ignore errors based on file descriptor, and I won't convince you 
that the Fatal module is a stupid hack.  Let's move on and get some 
actual code written.

--
Dustin Sallings