7a12a014   
   From: grahn+nntp@snipabacken.se   
      
   On Sun, 2010-02-21, David Schwartz wrote:   
   > On Feb 19, 3:18 pm, Jorgen Grahn  wrote:   
   >> I can't see [accept(2) may block after select]   
   >> as a positive thing.  For the rest   
   >> of the task I'm looking at, I can avoid blocking completely. It's just   
   >> this odd error case in the connecting phase (which I have been unable   
   >> to trigger in controlled experiments, by the way) which still is   
   >> unsafe.   
   >   
   > There is no way to avoid blocking completely with blocking sockets.   
   > You are wrong about it just being this odd error case. This odd error   
   > case is the only case you know of. The problem is all the cases you   
   > *don't* know of, or the cases introduced by the next version of the   
   > Linux kernel, or the next library your code is linked to.   
      
   Which would these cases be, then and where can I read about them in   
   the literature?  I haven't read as much as I perhaps should, but I   
   find it hard to believe that select/pselect/poll is unusable with   
   blocking, and that the books neglect to point it out.   
      
   I would have rejected your idea by now ... but you were right about   
   that accept(2) case.   
      
   /Jorgen   
      
   --   
     // Jorgen Grahn    O  o   .   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)