I totally digg ya but let me try to rephrase my point in a different way so maybe you'll understand what I am talking about.
.....
I still didn't get from your explanation the advantage of bucket mode.
Almost sounded like: If I render with progressive mode I need to wait 10 seconds until the render region clears out, but if I render with the bucket, I need to wait 10 seconds until the render region in the bucket clears out. So it's better.
Because 10 seconds are faster than 10 seconds. Or something :)
sounds almost like superstition :)
A similar case like this was some years ago on Steinberg forum. They changed the skin and UI colors for Cubase (audio multitrack app). And even though the engineers and programers themselves told the users the audio engine is the same, the users there contradicted themselves for days that the new version sounds better, others said no it sounds worse.. and so on)
Don't call something you can't understand superstition, it's just about explaining it simply enough for all to understand.
You're not thinking far enough. You're thinking 100% rendered on 100% of the region. That is the same with bucket and progressive and you are right.
But with bucket you don't need that, you
sometimes need only a small fraction rendered on 100% quality, in scene, to test materials.
It takes
10 seconds to
clear a region on
progressive for
64passess. For the same region, it takes just
1 second to render
10% part of that region on 64passes. By then you can maybe
already see, for instance, if your
glossiness map is okay and stop the render.... while on
progressive you still need to wait the
whole 10, maybe 9 seconds to clear the noise in the glossiness for the whole 100% of the image.
This demands for more small render regions that you need to do, the more you need to wait on parsing and building acc structure, so the whole process takes a lot longer.
Bucket mode should be updated to follow mouse ponter, so the checking would be even faster, without the need to do more render regions, and in my opinion should not be discarded as it seems - the lesser option in the adaptivity war.
Because, believe it or not, for some, bucket mode was
not about adaptivity at all. I usually rendered 1render pass on 64initial passess, didn't even know bucket adaptivity ever worked...
Someone mentioned IR, but I don't know what kind of hardware you guys use, but on some consumer desktop it's useless for quality control in a built scene, way too many slowdowns, only usefull on shaderballs, small scenes or extremely small regions on large scenes.
If still not clear, .... I don't know how else to explain it without doing some video or a diagram for you :)
You're all thinking "adaptivity war", and me and nkilar are thinking "workflow optimization"
Still, we'll see what 1.4 brings us in terms of everything written on this topic.