[QGIS-Developer] Broken 'randomness' for new layer styles?

[Nyall Dawson]

On 3 February 2018 at 19:17, Richard Duivenvoorde <rdmailings at duif.net> wrote:
> On 03-02-18 05:17, Nyall Dawson wrote:
>> Coincidentally, Mathieu and I have been speaking this week about
>> approaches to improve the random colors picked for layers. We came up
>> with a similar approach:
>>
>> - find a palette of ~100 colors which we think look good for layers,
>> and remove gross colors from this
>> - include this palette as an additional pre-installed color palette
>> within the standard QGIS install
>> - on startup, pick a random starting color from this list and a random
>> direction (forward or backward). When adding layers, grab this color
>> for the layer and then jump to the next/previous color in the preset
>> palette for the next layer added.
>> - (for 3.2 : allow users to pick which palette to use for new layer colors)
>>
>> How's this sound? I'd love to address this for 3.0 - our default
>> random color selection often picks yuck colors, and I think it'd be
>> great if 3.0 "magically" produced nicer looking maps for users without
>> any work required by them!
>
> Plan sounds great, but I think we should not try to do this for 3.0.
> The old 'random' colors are ok for now, there are really too much issues
> in current master to fix, which will affect average users.

The difference is - fixing the default color assignment is a FUN
project, and something I'm motivated to tackle before release.

On the contrast... if I have to look at any more Python binding
issues/memory leaks/object conversion code, I'm likely to end up in a
catatonic state and be no use to anyone!

So for me at least the choice isn't between fixing the random colors
vs fixing a crash - it's fixing the random colors vs not fixing them.

Nyall

>
> Yesterday I hit crashes (#18019), freezes (#18032) and failures to save
> as gpkg (#17899). It is better for the project (I think) if we clean up
> this kind of user experiences first :-)
> I know I started this color discussion, but still...
>
> I encourage everybody on this list to start using 2.99/3.0 and try to
> confirm or invalidate issues from the issue list. We really need to test
> for usability.
>
> Regards & thanks for all the hard work and great ideas people!!
>
> Richard Duivenvoorde