Fix missing symbol issue on Windows for filters::convolution and poisson4::bspline

[OgreTransporter]

Here is a new bugfix for problem #4082. This is a different approach than in PR #4151. Instead of undoing PRs #3972 and #3971, we use PCL_EXPORTS here. Many thanks to larshg who found the solution.

Closes #4082

[OgreTransporter]

@UnaNancyOwen, @SunBlack: Could you please also test this solution in your environments?

[SunBlack]

Looks good 👍

[kunaltyagi]

Thanks a lot @OgreTransporter

This looks much cleaner. I'll close the other PR in favor of this appraoch

[OgreTransporter]

I have built and run the test as debug and release with Visual Studio 2017 and 2019 on different computers. Works fine for me.

[UnaNancyOwen]

LGTM 👍

[kunaltyagi]

Haven't reviewed, but I love the 3000 removed lines 😄

PS: tests failing

[OgreTransporter]

Yes, I have made a mathematical approximation to the original data and then determined the new output data. In addition, I have also determined output data for the other tests. I did not add the data line by line to an array - as in the original test - but initialized it as a fixed array. Fewer lines and you can test comfortably.

The macOS build failed due to a signed/unsigned comparison. Nothing happens with GCC, under MSVC there is a compiler warning and under OSX it is an error.

[SunBlack]

The macOS build failed due to a signed/unsigned comparison. Nothing happens with GCC, under MSVC there is a compiler warning and under OSX it is an error.

The macOS build is configured with warnings as errors, during the other not (now) - that's why the build is failing only on macOS.

[kunaltyagi]

Code looks good, just some minor inline comments to improve readability.

You've mentioned that you copied the tests. If this refactoring didn't touch those tests, could you create an issue with the details of the tests that can be refactored too? Thanks for the work you've put in for this (these) PR

[OgreTransporter]

You've mentioned that you copied the tests. If this refactoring didn't touch those tests, could you create an issue with the details of the tests that can be refactored too? Thanks for the work you've put in for this (these) PR

Unfortunately the statement is no longer true. The test principle of generating and filtering inputs and comparing the outputs is still the same. But by approximating the data, new results have come out. So the original test has been renewed as well.

[kunaltyagi]

LGTM. Needs one more set of eyes before merge

[OgreTransporter]

d:\a\1\s\surface\include\pcl\surface\impl\surfel_smoothing.hpp(84): fatal error C1083: Cannot open compiler intermediate file: 'pcl_surface.dir\Release\mls.obj': Not enough space [D:\a\build\surface\pcl_surface.vcxproj]
LINK : fatal error LNK1257: code generation failed [D:\a\build\surface\pcl_surface.vcxproj]

This time it's not my fault ;-)

[kunaltyagi]

@SunBlack, @larshg Could you please give this a look over? Especially the tests. If everything is ok, let's merge this to master

[larshg]

I have build without PR: pcl_train_linemod_template failed with linking errors as expected.
Build with PR: pcl_train_linemod_template is build succesfully.
The modified and new unit tests runs successfully.

I haven't verified the approximation of the original data, nor reasoned anything about its correctness in general.

But it compiles and run. I have only launched the pcl_train_linemod_template, without any arguments. But I just assume it works as it hasn't changed per se.

[kunaltyagi]

Thanks @larshg

Let's merge this