yost@esquire.UUCP (David A. Yost) (02/15/90)
I'm writing a filter that has to work in real time with binary input, so I am calling reads(). If I call reads(), everything works, but extremely slowly because reads() defaults to reading one character at a time. So I call reads() with a larger number as the count argument. Problem is, reads() insists on waiting until all of those characters have come in before it will return. This is nice behavior for some situations but horrible for others, in particular my situation, where I need reads() to return with whatever it got. This leads me to suggest that reads() should have an additional argument, perhaps as in this spec: reads (f,i,j) Values for j are: 0 (the default) will not return until it gets all i characters or until end of file. 1 returns with a partial count rather than block. 2 returns with a 0 or partial count rather than block. Has anyone done anything like this? I'm really just asking for access to more of the normal functionality of the underlying UNIX system call. --dave yost yost@dpw.com or uunet!esquire!yost
kelvin@ASTRO.CS.IASTATE.EDU (02/15/90)
you might want to look at:
A Stream Data Type that Supports Goal-Directed Pattern Matching on
Unbounded Sequences of Values, in Computer Languages, Vol. 15, No. 1
(jan 1990) by Kelvin Nilsen
this describes one proposed solution to the sort of problems you mentioned.
unfortunately, i haven't yet gathered enough external funding and/or
academic rank to spend much time on development and distribution of
my real-time derivative of Icon, Conicon.
Kelvin Nilsen/Dept. of Computer Science/Iowa State University/Ames, IA 50011
(515) 294-2259 kelvin@atanasoff.cs.iastate.edu uunet!atanasoff!kelvin