On Fri, Jul 12, 2013 at 03:55:16AM +0900, Eric Wong wrote:
> Aaron Patterson <tenderlove / ruby-lang.org> wrote:
> > It occurs to me that this is pretty similar to IO#readpartial.  The main
> > difference being that readpartial automatically retries on EWOULDBLOCK
> > and raises an EOF error.  Maybe there should be a `try_readpartial` in
> > addition to the proposed `try_read_nonblock`?
> 
> It's always bothered me that IO#readpartial raises EOFError, too
> (especially when IO#read returns nil).  So yes, try_readpartial
> would be good, but try_*nonblock is much more important since
> EAGAIN is more common than EOF.

Ah, I was thinking that IO#try_readpartial would work exactly the same as
try_*nonblock, but with a buffer (no exceptions, no automatic retry).

> Also, I also wonder if there's a generic way for us to implement
> "expected exceptions":
> 
> - without needing to introduce additional methods
> - without allocating new objects for common exceptions
> 
> Perhaps similar to catch/throw...
> 
> Maybe:
> 
>   begin_expect(Errno::ENOENT)
>     # cause the _class_ Errno::ENOENT to be raised,
>     # not an instance of Errno::ENOENT
>     File.open("/possibly/non-existent/file")
>   rescue Errno::ENOENT => e
>     e == Errno::ENOENT # note the '==' (not '===')
>   end
> 
> Otherwise, we might end up needing try_open, try_stat, try_lstat,
> try_link, try_unlink, try_rename, etc to avoid racy/wasteful
> File.exist? calls...

I'm not sure.  The big thing for me about the try_* methods is that I
can use `loop` as the looping construct, not rescue / retry.

-- 
Aaron Patterson
http://tenderlovemaking.com/