Slow slicing of thin walls

[kefir-]

I'm currently trying to slice this stl, but so far it's taken 80 minutes CPU time on my quad core laptop. This is the slowest I've ever seen slic3r do, so I figure this must have hit a suboptimal part of the algorithm somehow. It's a wing profile, basically made up of thin walls. Here's the STL:

https://docs.google.com/file/d/0B947HPmfcUmTZW1MUGt5enlZN0E/edit?usp=sharing

If necessary I can share my settings later.

[alranel]

Yes, current implementation of medial axis is slow.

Can you test the thinwall-bpv branch, where the new one is being developed? The main purpose of it is to get more robust and cleaner toolpaths for single wall parts, but it should be also faster.

[kefir-]

Yeah, that was much faster. The first one is still processing after 90+ minutes CPU time, the thinwall-bpv branch took 8min20s. Still takes a while, but big improvement!! :)

[alranel]

@mesheldrake (thin wall specialist) will be happy to hear that ;)

[kefir-]

The thinwall-bpv branch has some peculiar ordering of the lines it puts down, but I guess that is stuff that may have been improved in master already?

[alranel]

@kefir-, thin wall logic is only in thinwall-bpv, so you might be noticing the logic that tried to make continuous paths by connecting thin walls to adjacent perimeter loops. I'll let @mesheldrake handle this as he knows the topic :)

[fehknt]

I'm so glad I found this issue. I had a job running for >2 days, before deciding that it just was never going to finish. Tried thinwall-bpv and it was done in <5 minutes. WOW. (It was an MRI brain scan surface reconstruction.)

[alranel]

@fehknt, wow, that's an impressive benchmark. I hope we can merge that branch soon!

[alranel]

@kefir-'s, object took 2 minutes and 43 seconds with default settings using the boost-medialaxis branch. Result looks clean and correct: