Use assimp to load STL files

[simonschmeisser]

Description

rviz currently uses home grown code for loading STL files which does not support ASCII format but only binary format. As discussed in #1364 this is probably for historical reasons, ie stl_loader predates usage of assimp for loading other file formats (most notably collada dae)

This PR removes STL_loader and uses assimp instead.

Remaining issues

• Loads a sample setup from ros-i/fanuc
• assumes all files to be collada and tries to read scaling info from stl
  [ERROR] [1554565767.625447979]: XML parse error [package://fanuc_cr7ia_support/meshes/cr7ia/visual/base_link.stl]: XML_ERROR_PARSING_TEXT
• Unit test fails to load stl files, most likely due to incomplete setup
• test for regressions on a subset of ros-i meshes
• discuss regression of last sample file (file contains more binary data than specified in header)

[simonschmeisser]

Quick note on the ascii test file included with rviz:
if contains end loop and end facet while typical files and the sample at wikipedia have endloop and endfacet (assimp seems to start a new facet instead of ending the current at the space between end and facet)

[simonschmeisser]

@rhaschke I had another look at this testcase and there seems to be no way to run this truly headless (unset DISPLAY). What works however is running it with
xvfb-run make run_tests_rviz_gtest_stl_loader_test Do you think this is a viable approach and if so do you know how to combine it with moveit_ci? I see two options, either telling catkin_add_gtest somehow to prefix xvfb-run to just this test or running the whole test within xvfb-run. (Starting Xvfb as a background process unfortunately lead to a segfault when running the test for me)

[rhaschke]

The test fails on the ROS build farm, which indeed doesn't provide a (virtual) X server. moveit_ci in Travis does. What we do in MoveIt in such a case is disabling the test, see here.
I'm puzzled though, why this PR is not tested on Travis as well.

[simonschmeisser]

I had the "WIP" set until a moment ago, maybe that stops travis from running

[rhaschke]

I had the "WIP" set until a moment ago, maybe that stops Travis from running.

Interesting. Looks like this disables Travis indeed. Didn't know so far...

As your PR changes ABI, we can only schedule it for Noetic.

[simonschmeisser]

Thanks for your hint for Xvfb!

I fixed ABI compatibility by re-adding the stl_loader class. I guess I should add a deprecated of some sort to it? Will this "dead-code" then be removed somehow?

I re-added the files using git checkout HEAD~4 filename, does this preserve history (after squashing)?

[rhaschke]

Yes, please mark the class (or its constructor(s)) as deprecated with a TODO comment when to remove it (in Noetic). Squashing creates a new commit based on the actual remaining differences, so yes, history will be preserved.

[simonschmeisser]

I added a deprecation note and comment. Please note that this will be quite noisy without #1367

[rhaschke]

In general looks good to me. @wjwwood, I would like to have your opinion as well before merging.

[simonschmeisser]

thanks for merging @rhaschke