Re: about lposix

[Rici Lake <lua@...>]

On 17-Oct-06, at 9:08 PM, Mike Panetta wrote:

> Only thing I am worried about is if this library only handles the 
> least common denominator (which seems to be windows, atleast in one of 
> the cases you mention below) then the library will not truly solve the 
> problem of different people going off and doing their own thing, 
> because there are people out there that need the 'extra' features 
> (like me, and I am very sure I am not the only one).

Yes, I think that would be a mistake.

Posix, as has been mentioned, does not require every single feature to 
be implemented. However, it defines how features should be implemented 
if they exist.

To take an example which came up on IRC today, it is not clear how to 
implement symlinks on Windows. On the other hand, it is not guaranteed 
that a given symlink can be created on a given filesystem. So anyone 
trying to create a symlink needs to check the error return. If a 
particular OS didn't implement symlinks at all, it would be legitimate 
for the implementation of <whatever>.symlink() to simply return an 
error code, perhaps EPERM (which is apparently what Linux returns if 
you try to create a symlink on a filesystem which doesn't support 
them.)

Speaking of EPERM and friends, I'm of the opinion that "posix-like" 
error codes should be returned as strings ("EPERM") rather than 
whatever number happens to be the value of EPERM on the particular os. 
The strings are a lot easier to deal with in script code, and there's 
no point I can see to defining errno.EPERM as some number rather than 
just having the API return "EPERM".