Re: [BK PATCH] One strdup() to rule them all

[Dan Aloni <da-x@gmx.net>]

On Mon, Aug 25, 2003 at 06:37:45PM +0200, J?rn Engel wrote:
> > +/**
> > + * strdup - Allocate a copy of a string.
> > + * @s: The string to copy. Must not be NULL.
> > + *
> > + * returns the address of the allocation, or NULL on
> > + * error. 
> > + */
> > +char *strdup(const char *s)
> > +{
> > +	char *rv = kmalloc(strlen(s)+1, GFP_KERNEL);
> > +	if (rv)
> > +		strcpy(rv, s);
> > +	return rv;
> > +}
> 
> My gut feeling is always afraid when something "must not be NULL",
> someone will ignore this and Bad Things (tm) happen.  Is strdup ever
> used such performance critical code that the extra check would hurt?

There are two reasons while it shouldn't have a NULL check. One,
persistency: the other str*() functions don't do this sort of check.
Two, for general uses like that:

     new_name = strdup(name);
     if (!new_name)
         goto allocation_failed;
	 
With this check, NULL would be returning from strdup() either because 
name == NULL or when the allocation fails. You cannot simply pass that 
information through its return value and the caller would need to 
check whether name == NULL by itself which should have been done anyway.

Passing NULL to strdup() is a bug, which would have NOT show as an
Oops if you add this check, and that is bad. Maybe it would be worth 
to add a BUG_ON(s == NULL) instead, and perhaps also add this to the 
other str*() functions, but that is a different patch.

-- 
Dan Aloni
da-x@gmx.net