TestParticleAtlas crash, assert failure

[0joshuaolson1]

Here's what little I have so far. My last run I had TestParticleAtlas print the duplicates, sorted in a copied vector like hasDuplicates does (the first of which is the failing assertion):

112 40000
114 40000
116 40000
118 40000
316 40000
318 40000
320 40000
322 40000
519 40000
521 40000
523 40000
525 40000
527 40000
725 40000
727 40000
729 40000
928 40000
930 40000
1120 40000
1122 40000
1124 40000
1126 40000
1128 40000
1130 40000
1132 40000
1136 40000
1328 40000
1330 40000
1332 40000
1334 40000
1336 40000
1338 40000
1340 40000
1342 40000
1536 40000
1539 40000
1541 40000

There are several groups of different sizes where every other one is a duplicate (out of 40000). When it doesn't crash (always before printing anything), it's always these indices. I changed genPoints in unittest/util.h to be pseudorandom, and the groupings are similar. I'll post this after I figure out more about the crashes.

By the way, do you have any ideas how to prevent AppVeyor from rebuilding the dependencies and ALL of openvdb? I tried caching folders, but the compile-download-run cycle is still over twenty minutes (log). Does CMake need to be told something? Are subfolders not being cached?

[caron]

If you look at my repo you will see my attempts to cache the dependencies but I haven't finished it. Honestly I don't know the best way to improve the build time right now. I was thinking I might make a new repo which builds all deps and then packages everything for which the openvdb project consumes. That will take more time but could simplify the appveyor.yml a lot. BUT that doesn't solve the long build time for the unittest... If I were you, while debugging you could disable most of the tests by modifying the CMake to not glob all files but just put the ones you need...

https://github.com/0joshuaolson1/openvdb/blob/master/openvdb/unittest/CMakeLists.txt#L3-L4

[0joshuaolson1]

I used your caching/IF NOT EXIST stuff as a guide, and I'll try changing the glob strategy. I'm in no hurry though. I'm gonna see what the tests do for me on Linux.

NOTE: Gave up on Linux for now. Not going to waste time building dependencies since no one watching the forum has replied with what kind of builds (static/dynamic) the dependencies should have.

[caron]

Ah, yes... that was very very incomplete and only saved time on the download of source code and/or binaries for some packages. It didn't actually choose to skip building if the cache was restored. I will need to make a more complex appveyor.yml file or make a new external build script.

[0joshuaolson1]

That would make recompiling sooo much better!