hi, basically karab44 wrote skinify 0.89 on windows and wrote 0.90 on linux, causing a platform depend somehow?

that said, 0.89 is claimed not to work on linux either: 0.89 works win, 0.90 works with linux

I back tested 0.90 and issues exist in older builds, i forward ported 0.89 and it's good on rc2 windows
I only can test on win.
the solid shape code was changed between versions in the addon, the solid shape button causes the failed geometry especially in hands.

The animal rigify skin fine, rigify rigs fail, manual bastioni rigs fail with skinify: improper skinning for hands and arm taper towards wrist.
*so i could rule out if it was issue with rigify i tested with manual bastioni human lab and had same issues: rule out skinify interaction with rigify was broken.
I also checked bf commit logs, not really anything there to cause this between my test versions 9cd6b03 and rc2
i bisected to a possible commit, but then, the forward porting worked and backporting before the possible commit still failed, so i ruled that out.
here's diff of both versions: https://developer.blender.org/rBA8a3e171dabe636e089ed1f7beb2c09a101ab2afb

thanks

The report has too many steps, at least narrow this down so we can tell if it's Blender or the script.

It looks like the code from rBA8a3e171dabe636e089ed1f7beb2c09a101ab2afb depends on order in dictionary, I've asked @Albert (karab44) on IRC to try using PYTHONHASHSEED for deterministic hashing, but didn't hear back on this.

When looping over a dictionary, you cant expect predictable outcome, try putting calls to dict.keys/items/values inside a sorted(...).

Sorted dictionaries only work if memory locations arent used as keys, so if this doesn't help, best prepare the data from the script, before modifiers run as a static blend file, then we can run the modifier on it and see if its giving different results depending on the system.

@Campbell Barton (campbellbarton) can not confirm script bug at the moment. PYTHONHASHSEED and DETERMINISTIC_BUILD does not help nor OrderedDict.

So far provided cross platform solution. Time will show is it final fix or workaround.

Best Regards!
Albert

including missing file from previous post

This seems like a fragile workaround - does it work with 32 bit systems? seems like someone using BSD's will get some empty mesh or error too.

Would rather see this simplified into a static blend file with that a developer can check.

hi, I will supply a .blend, after talking with @Albert (karab44) , it will have the 2 skinify scripts (this revision and the previous broken revision) in the text editor and 2 rigs for testing.
I'll set it up tonight and supply tomorrow.
thanks

rB2ca1f297486804dcdb90c376e68dce656a73c726 may fix the issue.
rBA8a3e171dabe636e089ed1f7beb2c09a101ab2afb will need to be reverted first.

hi, @Albert (karab44) can you test this please.
@Campbell Barton (campbellbarton)
to test this addon is simple.
create a rigify rig "Human" (Meta Rig)
This result is from RC2 on win64 with current version of skinify, which is tested as working on the 3 major platforms. win, osx, linux 64bit.
run skinify with the following settings:

A note:
this was a fun addon with many uses, proxy mesh for characters, test mesh for animations, quick set up, easy enough to use for all user levels.
we tested then found the cross platform issue ourselves, then in irc found testers, I'm happy to add a warning to the current version that "may not work as expected".
The issues is the hand mesh, and issue in old addon was the platform compatibility. We have solved these issues in the version 0.9.9 of skinify addon. Whilst the code may not be perfect, the times where it may fail are limited to hand rigs and the addon will work as expected in the majority of cases.

Can we just keep this working and add a warning.
Thanks.

hello!

@Brendon Murphy (meta-androcto) no worry, all good.

New results are invalid just because the result was received on the setup and calculations on the 'outdated/invalid' data(triggered by the cpp code update). The upcoming updated script should work fine for all platforms with the script data based on linux idx input only.(win and darwin input removed)

So the current script requires to update the idx data for all platforms to be set exactly the same as current idx settings for the linux OS .
However this should be additionally tested for all platforms.

I've been waiting with the script update untill all cross platform builds are being available for all architectures so expect the script update based on fixed solution soon.

Best Regards!
Albert

@Brendon Murphy (meta-androcto)
rB2ca1f297486804dcdb90c376e68dce656a73c726

it's not an issue, it's the solution but the script relies on outdated/invalid input.

New update will no longer be platform dependent anymore. I'll update the script on Monday (2moro)

@Albert (karab44) note that master now uses KD-tree for remove doubles, see: rB459d76ec5106a949f85c121a3d977a65af560f4c rB4d148471b66a7f7c2533e3d1d2b037fd838b9bc4

@Albert (karab44) - is this report just waiting on you updating the script for changes to Blender?

@Campbell Barton (campbellbarton) yes, waiting on @Albert (karab44)

fresh and clean

Pep 8 Cleanup ( version 0.11.0)