Re: [PATCH] compat: Add another rudimentary poll() emulation

On Thu, May 27, 2010 at 4:32 PM, Marko Kreen [off-list ref] wrote:

On 5/27/10, Erik Faye-Lund [off-list ref] wrote:

quoted

On Thu, May 27, 2010 at 4:11 PM, Marko Kreen [off-list ref] wrote:
 > On 5/27/10, Erik Faye-Lund [off-list ref] wrote:
 >> On Thu, May 27, 2010 at 3:58 PM, Marko Kreen [off-list ref] wrote:
 >>  > On 5/27/10, Erik Faye-Lund [off-list ref] wrote:
 >>  >> On Thu, May 27, 2010 at 3:29 PM, Marko Kreen [off-list ref] wrote:
 >>  >>  > On 5/27/10, Erik Faye-Lund [off-list ref] wrote:
 >>  >>  >> On Thu, May 27, 2010 at 1:00 PM, Erik Faye-Lund
 >>  >>  >>  [off-list ref] wrote:
 >>  >>  >>  > On Thu, May 27, 2010 at 12:10 PM, Jonathan Nieder [off-list ref] wrote:
 >>  >>  >>  >> Implement the subset of poll() semantics needed by git in terms of
 >>  >>  >>  >> select(), for use by the Interix port.  Inspired by commit 6ed807f
 >>  >>  >>  >> (Windows: A rudimentary poll() emulation, 2007-12-01).
 >>  >>  >>  >>
 >>  >>  >>  >
 >>  >>  >>  > A possible problem with this approach is that the maximum number of
 >>  >>  >>  > file descriptors poll can handle limited by RLIMIT_NOFILE, whereas the
 >>  >>  >>  > maximum number of file descriptors select can handle is limited by
 >>  >>  >>  > FD_SETSIZE.
 >>  >>  >>  >
 >>  >>  >>  > I don't think this is a big problem in reality, though - both values
 >>  >>  >>  > seem to be pretty high in most implementations. And IIRC git-daemon is
 >>  >>  >>  > the only one who needs more than 2, and it doesn't even check
 >>  >>  >>  > RLIMIT_NOFILE.
 >>  >>  >>  >
 >>  >>  >>
 >>  >>  >>
 >>  >>  >> To be clear: I think this strategy is the best option (at least for
 >>  >>  >>  non-Windows, where select() might be our only option).
 >>  >>  >>
 >>  >>  >>  But perhaps you should include a check along the lines of this:
 >>  >>  >>
 >>  >>  >>  if (nfds > FD_SETSIZE)
 >>  >>  >>         return errno = EINVAL, error("poll: nfds must be below %d", FD_SETSIZE);
 >>  >>  >>
 >>  >>  >>  Just so we can know when the code fails :)
 >>  >>  >
 >>  >>  > Well, per your own FD_SET example, the FD_SETSIZE on windows
 >>  >>  > means different thing than FD_SETSIZE on old-style bitmap-based
 >>  >>  > select() implementation.
 >>  >>  >
 >>  >>  > On Unix, it's max fd number + 1, on windows it's max count.
 >>  >>  >
 >>  >>
 >>  >>
 >>  >> Are you sure this applies for all Unix, not just some given Unix-y system?
 >>  >
 >>  > Not sure.  Just pointing out that the above check is not
 >>  > universal enough.
 >>  >
 >>
 >>
 >> Isn't it? How could one possibly pass more than max fd number + 1 file
 >>  descriptors, since they start at 0? I guess one could specify a given
 >>  fd more than once, but that'd be kind of redundant... and also very
 >>  unlikely in our case ;)
 >
 > Pass one fd with value 70 it.  Check returns error, although
 > everything would work.
 >


No, not with the check I posted. I checked nfds, not the value of the
 fds themselves. nfds is clearly the size of the fds array -- if it
 wasn't, it'd be impossible for the poll-implementation to know how big
 the array is!

Ah, ok.  I though this is the nfds for select().  In this case it can
still fail, as nfds=1 for poll() can still go over FD_SETSIZE for
select() if the fd value is big enough.  On non-windows, that is.

I don't quite get where one would get a fd that's value is FD_SETSIZE
or above from. I mean, FD_SETSIZE would have to be the maximum value
of a file descriptor, or else you risk open() or socket() returning
file descriptors that cannot be select()'ed. Which would just be
insane, no?

-- 
Erik "kusma" Faye-Lund