From: Tanaka Akira <akr@...17n.org>
Date: 2004-07-23T19:36:58+09:00
Subject: Re: sysread and buffered I/O

In article <gGSdnSk4oLg_DZ3cSa8jmA@karoo.co.uk>,
  "daz" <dooby@d10.karoo.co.uk> writes:

> I thought Hal's read_avail (perhaps readavail) was a fair description, but ...
>
> "readpartial may return data which is not available when readpartial is called"
>
> Then 'readpartial' is magic ?-)  (Returns unavailable data)
> I know ... I misunderstand :(

readpartial blocks in such case.

% (sleep 1; echo abc; sleep 1; echo def) | ./ruby -e 't1 = Time.now; p STDIN.readpartial(4096); t2 = Time.now; p t2-t1'
"abc\n"
1.001787

In this case, readpartial blocks because there are no data at first.
After 1 second, "abc" is sent over the pipe and readpartial reads and
returns it.

So, the name read_avail is not accurate.

>  read_immed ?
> (readimmed is *very bad* without the underscore -- weird ! )
>
>  read_direct ?  (probably not good)
>
>  read_instant, readinst, readsnap ?  (at this instant in time)

How about them, matz?

> I think the problem with the name 'readpartial' may be that if *all*
> data is available, and 'readpartial' reads it all, then it has "failed"
> because it has performed 'read', *not* 'readpartial'.   :->

Yes.

> 'readchunk' gives too much emphasis to the chunk, IMHO.
>  readchunk(256) looks too predictable.
> If I wanted to read from <ios> in chunks, I would try that before
> reading the docs.

Also, several formats such as PNG defines "chunk" in their spec.
-- 
Tanaka Akira