Regression tests and CGAL versions
My recent pull request builds and passes all tests with the Travis server.
"Yay!" I thought.
I thought I would do a sanity check and run tests on latest openscad/master.
If I run the tests on my own systems, I get 70-72% passing (depending on
which system). Some of the errors appears to be alignment (not centered),
some are rendering (edges appear different), some are CGAL assertion
failures and no output at all.
It appears that Travis CI is using CGAL 3.9, the CGAL package I have on my
Linux Mint systems is 4.2. I don't know if it is possible to get the log
output from Travis?
4.2 is dated April 2013, 3.9 is Sept 2011. The latest Ubuntu has 4.5.
I'm not sure how it should be decided which is the "best" version, but I
guess the version Travis has is not the version most people are using, and
it appears to make a significant difference.
--
View this message in context: http://forum.openscad.org/Regression-tests-and-CGAL-versions-tp13312.html
Sent from the OpenSCAD mailing list archive at Nabble.com.
On 07/26/2015 10:42 PM, bobc wrote:
My recent pull request builds and passes all tests with the Travis server.
"Yay!" I thought.
I thought I would do a sanity check and run tests on latest openscad/master.
If I run the tests on my own systems, I get 70-72% passing (depending on
which system). Some of the errors appears to be alignment (not centered),
some are rendering (edges appear different), some are CGAL assertion
failures and no output at all.
That's strange, for a normal "ctest" run, I'm getting this on master:
99% tests passed, 4 tests failed out of 922
Total Test time (real) = 712.30 sec
The following tests FAILED:
462 - opencsgtest_issue1215 (Failed)
764 - throwntogethertest_issue1215 (Failed)
908 - openscad-camtrans-viewall_camera-tests (Failed)
910 - openscad-camortho-viewall_camera-tests (Failed)
report saved:
Testing/Temporary/linux_x86_64_gallium-0.4-on-nve7_caiu_report.html
=> http://files.openscad.org/tests/linux_x86_64_gallium-0.4-on-nve7_caiu_report.html
I'm not sure why the two camera tests fail, the other two seem to be some
z-fighting issue.
Could you upload the test output somewhere?
It appears that Travis CI is using CGAL 3.9, the CGAL package I have on my
Linux Mint systems is 4.2. I don't know if it is possible to get the log
output from Travis?
The run log is at https://travis-ci.org/openscad/openscad/builds/72729137
(link is at the green check mark on the pull request / commit).
Normally the test results should be uploaded to http://files.openscad.org/tests/
but it seems that stopped working some time ago.
I'm not sure how it should be decided which is the "best" version, but I
guess the version Travis has is not the version most people are using, and
it appears to make a significant difference.
I think there were some critical fixes somewhere around the 3.6 version
but otherwise the CGAL version did not seem to make much of a difference.
I'm currently using CGAL 4.6 on my Debian system.
ciao,
Torsten.
Ok, I see the travis log, but it only has the pass/fail line.
Ah, I just realised I goofed, I ran All which includes known issues. But I
think I still get some failures, it passes 845 tests.
I'll rerun it tomorrow and post the report.
--
View this message in context: http://forum.openscad.org/Regression-tests-and-CGAL-versions-tp13312p13315.html
Sent from the OpenSCAD mailing list archive at Nabble.com.
Generally I want to comment on CGAL stability:
When I create an openscad script with medium or big size, Openscad is
quite good
in calculating the final shape. I like its quality.
However, when I import existing stl files into openscad because I want to
alter them
(cutting, adding, etc ), I got a very hard time.
Most of the results: it cannot evaluate and the result is nothing.
Sometimes it helps, pushing the STL file to netfabb first, then try again,
but overall success rate is no bigger than 50%.
Is this the actual quality of CGAL or is it just a stupid error on my side
?
Guenther
On Sun, 26 Jul 2015 15:03:15 -0700 (MST), bobc
bobcousins42@googlemail.com wrote:
Ok, I see the travis log, but it only has the pass/fail line.
Ah, I just realised I goofed, I ran All which includes known issues. But
I
think I still get some failures, it passes 845 tests.
I'll rerun it tomorrow and post the report.
--
View this message in context:
Sent from the OpenSCAD mailing list archive at Nabble.com.
OpenSCAD mailing list
Von: "Guenther Sohler" mail@guenther-sohler.net
However, when I import existing stl files into openscad because I want to
alter them (cutting, adding, etc ), I got a very hard time.
Most of the results: it cannot evaluate and the result is nothing.
Sometimes it helps, pushing the STL file to netfabb first, then try again,
but overall success rate is no bigger than 50%.
Is this the actual quality of CGAL or is it just a stupid error on my side?
The issue with STL is that it's a very simple way of encoding objects.
It duplicates information about points which could lead to floating
point rounding issues both at the save and the load implementation.
There is a huge percentage of STL files around that have really bad
issues.
CGAL is very picky about the geometry as the CSG operations normally
expect 2-manifold objects. One issue is that 2-manifold is a bit
more strict than most practical things require (e.g. 2 cubes sharing
one edge). I'm not sure how other CSG engines behave regarding this
but I suspect it's not much different.
The biggest issue remaining is STL files that have zero-faces. This
refers to triangles that have no area due to all 3 points being on
one line (or even all 3 points are identical).
The problem with those is, that it's impossible to create a face
normal vector to determine the direction the face is pointing to.
As workaround it often helps to fix zero faces using MeshLab.
https://en.wikibooks.org/wiki/OpenSCAD_User_Manual/FAQ#Why_is_my_imported_STL_file_only_showing_up_with_F5_but_not_F6.3F
So it's not simply CGAL which is bad and it's certainly not a user
error.
ciao,
Torsten.