new working set icons... not unique?

Matthew Woehlke matthew.woehlke at
Wed Dec 18 22:55:20 GMT 2013

On 2013-12-18 17:23, Sven Brauch wrote:
>> I was sufficiently curious to play around with it :-). Attached is the
>> script (using PySide, can likely trivially adapt to PyQt) I whipped together
>> to see what it would look like. The 3x3 grid isn't terrible. If you'd be
>> willing to lose the clear separation of cells, 8x8 with 128+ steps could
>> also be interesting. (IIRC the algorithm I was reading about only uses a
>> grid of about 7x9. You can get away with a much lower grid size if the cell
>> color isn't constant.)
> Haha, nice. Yes, I'm surprised the 3x3 grid is actually quite okay.
> Still, I see the same problem here ... the number of 16x16 icons which
> are distinguishable and rememberable are simply too limited to
> generate them in a way such that random selection works without
> clashes. The current algorithm doesn't have a lot of clashes, but as
> you see, it's enough to make people unhappy ;)
> The more possibilities your algorithm uses, the more formally
> different icons you can generate of course, but that goes at cost of
> recognizable and rememberable icons... and I tried to find a balance
> between them. I failed, in that there's not enough options to make it
> work without clashes. But I don't think adding more options works as a
> solution.

Give yourself some credit :-). As far as distinct options, given the set 
I'm currently looking at I can clearly see that there are no fewer than 
31 "pretty-high-quality" options (15 bit patterns times either red of 
blue, plus all dark). The failure is in the "apparent reuse" (or 
whatever you want to call it).

I can see where 3x3 might have a detrimental effect on memorability. 
What's interesting is that memorability probably starts to go *up* after 
a certain point once you get to where more abstract patterns can emerge 
(thinking again about the ASCII visual fingerprint problem). I wonder if 
anyone has done studies on that? :-) (Just being amused here, not making 
suggestions for KDevelop.) Relatedly the "noise" from the color I 
suspect goes away at higher grid sizes as it becomes a more meaningful 
indication of population density. IOW, once it starts to look like a 
Rorschach test rather than static, it becomes more memorable ;-).

(Note: my previous thought as far as trying 3x3 was still to keep a 
single color.)

> Overall, here's my suggestion: I will write something which either
> stores the icons or does a different clever thing so that we're
> guaranteed to not get indistinguishable icons any more. That'll take a
> few days and then we can talk about how memorable they are again
> maybe.

Yes, I think that much is desirable. Thanks again.


More information about the KDevelop mailing list