Commit c37ef2f probably introduced few problems.
at https://github.com/alexrj/Slic3r/blob/master/lib/Slic3r/GCode.pm#L220, role is not property of $loop, so slicer fail with if --randomize-start is enabled.
with --no-randomize-start, extrude_loop could be called with empty ExtrusionLoop (about 8-th call with test-squares.stl), failing nearest point lookup on https://github.com/alexrj/Slic3r/blob/master/lib/Slic3r/GCode.pm#L226 :
terminate called after throwing an instance of 'std::out_of_range'
After testing for empty ExtrusionLoop, it generates gcode, but perimeters on second layer are missing.
Files here: https://drive.google.com/folderview?id=0B9HM4UOKDQBLYWZNcWRFa3ZscG8&usp=sharing
Good, thank you: more unit tests are needed.
Second bug is caused by the known Clipper bug about open paths with coincident endpoints. Our workaround is not catching some cases...
No tests were covering randomize-start, which was not working anymore…
… after recent ExtrusionLoop refactoring. #2028
Failing test cases for Clipper bug returning empty result set. #2028
Actually, it looks like a distinct Clipper bug. I isolated it into a couple failing test cases and reported it to upstream maintainer: http://sourceforge.net/p/polyclipping/bugs/96/
Disable test until Clipper bug 96 is fixed #2028
Updated Clipper to r467 but it doesn't seem to fix #2028 yet
Revert "Updated Clipper to r467 but it doesn't seem to fix #2028 yet"
This reverts commit a62457d.
This is now fixed in Clipper.
Update Clipper to r468. #2028
Revert "Update Clipper to r468. #2028"
This reverts commit 6a64395.
Can confirm this occurs on Lubuntu 14.04.01 with latest git version. Reverting to version 1.1.7 fixes the problem.