Hi, Romano,

Romano Paolo Tenca wrote:
> 
> > I should have pointed out that the modified algorithm below
> > splits the group number from the remainder of the line;
> > therefore, some post-processing might be needed to re-attach
> > the three-digit prefix to the data (unless, of course, the
> > subsequent processing could use the bucket index for that
> > purpose).  So the time comparison may not be comletely fair.
> 
> This is a little true for all the test we made, someone use
> read/lines, someone read, someone ends with a block of line,
> someone with a block of blocks. More "exact" tests should be
> made with a file on disk and stop with a file on disk.
> 

Or else perform all tests beginning with a block of lines as
obtained from input (or the previous process) and ending with a
block of properly-orginazed lines.

That way we'd have comparisons of only the block rearrangement,
and not mix in I/O overhead (which would likely vary from box to
box anyway).

>
> A note, this does not work as one expect under 1.2.8:
> 

I was running under Core 2.5; since View is not (yet?) supported
on Mac OS/X, I'm sticking with the version that I can actually
use on all platforms I regularly use.

-jn-
-- 
To unsubscribe from this list, please send an email to
[EMAIL PROTECTED] with "unsubscribe" in the 
subject, without the quotes.

Reply via email to