Re: [dev] Local vs global variables

On Fri, Nov 06, 2009 at 09:50:05PM +0100, pancake wrote:
>Looks hard to me to think that plan9 is doing kernel2user memory
>copies to pass a errno string...

Kernel to user memory copies are cheap, especially when the
string is <128b. Syscalls are more expensive (but still cheap on
modern systems). None of this really matters much, though, given
that errors are exceptional condidtions. They happen relatively
rarely.

>Errno in unix is not for kernel land. But kernel wrappers in libc do
>the task to fill errno corretly. But at te end.. Errno is just a int,
>not a string. Why don't just change a pointer?
>
>In unix there's errstr for this and perror which is a funny spanish
>word that means dog-r :P
>
>But the point is.. Why are the negative numbers not used for this?
>Any error for a syscall should return a negative value that should be
>used as an index in an array mapped in the process space by the
>kernel. Like the vdso in Linux.
>
>That will be a good implemention for errno :)

Because Plan 9 doesn't do error numbers, it does error strings.
Even process exit values are strings rather than numbers. It
has a lot to do with the fact that Plan 9 is a distributed
system that has to work with a lot of heterogeneous servers
which may have any number of unusual error conditions. It also
doesn't do localization, which is another not wholly unrelated
matter.

-- 
Kris Maglione
Just because the standard provides a cliff in front of you, you are
not necessarily required to jump off it.
	--Norman Diamond