Why does get_lim_data ignore RLIM_INFINITY?

Why does get_lim_data ignore RLIM_INFINITY?

[Ken Brown]

Paul,

I'm puzzled by one of the changes you made to vm-limit.c:get_lim_data() 
in bzr revision 111152.  Specifically, it doesn't seem right to ignore 
RLIM_INFINITY as in the following comment:

+  /* Set LIM_DATA to the minimum of the maximum object size and the
+     maximum address space.  Don't bother to check for values like
+     RLIM_INFINITY since in practice they are not much less than 
SIZE_MAX.  */

According to POSIX, "If a call to getrlimit() returns RLIM_INFINITY for 
a resource, it means the implementation shall not enforce limits on that 
resource."

So I think this fragment of the original code shouldn't have been removed:

if (rlimit.rlim_cur == RLIM_INFINITY)
  lim_data = -1;

Ken

Re: Why does get_lim_data ignore RLIM_INFINITY?

[Paul Eggert]

On 03/19/2013 08:03 PM, Ken Brown wrote:

> So I think this fragment of the original code shouldn't have been removed:
> 
> if (rlimit.rlim_cur == RLIM_INFINITY)
>  lim_data = -1;

If rlim_cur == RLIM_INFINITY, then in practice rlim_cur is either
greater than SIZE_MAX (in which case the surrounding code will
treat it as if it were SIZE_MAX, which is the same as (size_t) -1),
or it is so close to SIZE_MAX that Emacs won't care about the difference.
In either case, the above if-statement isn't needed.

It's true that POSIX doesn't guarantee this practical property of RLIM_INFINITY,
but it's also true that various POSIX implementations have squirrelly
nearly-RLIM_INFINITY values such as RLIM_SAVED_MAX that the code needs
to work with, and the hassle of dealing with this variation
"correctly" is probably more trouble than it's worth, since (unless
I'm missing something) the code works as-is.