Eric Kow <[email protected]> writes:
> On Wed, Feb 24, 2010 at 00:27:27 +0100, Petr Rockai wrote:
>> Wed Feb 24 00:15:55 CET 2010 Petr Rockai <[email protected]>
>> * Adaptively choose display units.
>
> Oh, sorry! I got confused. Read email, Eric; don't skim!
>
>> PS: With the patch, the output looks like this:
>
>> ============== ================= =================
>> darcs-2.3 darcs-2.3.99.2
>> ============== ================= =================
>> (un)revert mod 99.8ms d=0.8ms 94.7ms d=1.6ms
>
> Yeah! It's good that we do this, but one thing we need to be careful
> about is to show the same time unit across the row so that we don't
> display confusing things like 700ms against 1s. I mean, people are
> smart enough to do the conversion, but when you're staring at these
> columns of numbers, you really want to make things as easy as possible.
I find it better to have 10s d=25ms than 10s d=0.0 ... also easier on
the eyes for the more usual cases like 3s d=420ms.
> So there ought to be some way for the table code to override or dictate
> the time unit being used. Also, the way I did the side-by-side merging
> of variant tables is kinda stupid. We would probably have an easier
> time at if we didn't build intermediate tables and just mashed the rows
> together first (and then picked the smaller timeunit across the whole
> row).
Another option might be to say
============== ========= ====== ============== ======
darcs-2.3 stddev darcs-2.3.99.2 stddev
============== ========= ====== ============== ======
(un)revert mod 99.8ms 0.8ms 94.7ms 1.6ms
...and maybe strip ^darcs- from the binary names, like this
============== ======= ====== ========== ======
2.3 dev 2.3.99.2 dev
============== ======= ====== ========== ======
(un)revert mod 99.8ms 0.8ms 94.7ms 1.6ms
> Other comments are that (a) it may not be necessary to display the
> deviation units if we always use the same one and (b) it may be useful
> to display the sample size (ugh, that's getting wide again).
Well, if we insist on same units, that still poses problem for things
like 3s, where the deviation would get very little precision (1 digit at
most). Maybe it would also make sense to get the numbers formatted with
at most 1 decimal place, instead of exactly 1, although that lose
decimal point alignment.
For sample, maybe we can encode it differently, say ?22s for <6 samples,
~22s for 6-20 and just 22s for over 20? That only adds one character and
only for the "dubious" cases. This can then go to the legend.
Yours,
Petr.
_______________________________________________
darcs-users mailing list
[email protected]
http://lists.osuosl.org/mailman/listinfo/darcs-users
