Re: Multi-threading in sort(or core-utils)

[James Youngman]

On Sat, Jun 28, 2008 at 4:27 PM, Bo Borgerson <address@hidden> wrote:
> Hi James,
>
> Each sorter's portion of input is delineated along line boundaries as
> detected by the main buffer-filling routine.  I don't think any
> multi-byte character set problems should have been introduced.

I see.   I agree, then.

> What type of issue specifically concerns you?  I'm going to start
> putting together some tests soon, and I'd like to include a test case
> that would exercise the type of bug you have in mind.

Given that you had done away with the feeder, I had assumed that you'd
just set things up so that for N sort workers on an M byte input,
worker i would start at the next line boundary after byte offset
(i*M/N).   I was essentially wondering how you could accurately
determine the actual line boundaries in a multibyte file.    But I now
understand that you are not doing things that way.

James.