SERVER: read(conn,1) is a pickpocket

• Date: Sat, 21 Oct 1995 03:04:18 PDT
• From:
  Alex Stewart <riche@crl.com>

(or should I say "picksocket".. anyway..)

I was just looking over some of the 1.7.9 server code, and noticed something 
which I thought I'd bring up..

It appears (and I've tested this too), that using non-suspending-reads 
(read(conn, 1)) on a network connection under the right conditions can 
actually steal input from a currently (suspended) reading task on that 
connection.  Old-style read calls when used on a connection which was already 
being read from would throw E_INVARG, but the non-suspending read stuff 
apparently doesn't check for this condition.

I'm trying to decide whether this is actually a good thing or a bad thing, 
actually, but my inclination is that it's probably a bad thing, as it 
introduces some potential unpredictabilities into the issue of net IO..  I 
thought I'd ask what other people thought, tho, before putting together a 
patch for it (which would be fairly minor)..

-R
-------------------------------------------------------------------------------
     Alex Stewart - riche@crl.com - Richelieu @ Diversity University MOO
                         http://www.crl.com/~riche/
            "Difficult answers lead to intelligent questions."

• Prev: Release 1.7.9 of the LambdaMOO server
• Next: Switching from Object to Byte Based Quota
• Index: MOO-cows
• Thread: MOO-cows