LaserScan does not get display in 1.10.15

[jihoonl]

I confirmed that 1.10.14 displays laser scan properly. It seems like transform is not applied.

Version 1.10.14

Version 1.10.15

[wjwwood]

@jihoonl can you provide a bag file for reproducing this?

[wjwwood]

@tfoote please hold the sync of hydro if you haven't already done it.

[wjwwood]

This change seems to be the most likely candidate: 9d5c5a8

@vrabaud have you noticed this or have you seen laser scans working with your patch?

[mikeferguson]

Just checked out latest source to check this out -- and found that if you set color transformer to Intensity, the points don't get transformed -- they just end up at the origin (set the points to very large, and you can see them sitting there). If you select flat or axis color, things work for me.

Tested on a recent robot, and also on this bagfile which you can use: http://fergy.me/slam/2010-10-29-armadillo-cci.bag

[mikeferguson]

And by latest source, I mean 1.10.15 (I had a whole different game of fun when I forgot to switch to the hydro branch, cause the indigo branch builds but doesn't run under hydro)

[jihoonl]

I can also see that correct points if I set color transformer as other than intensity.

[wjwwood]

I can confirm that this is due to 9d5c5a8:

The top left rviz has 9d5c5a8 reverted and the bottom right one does not.

[wjwwood]

This seems like a candidate:

1779817#diff-406fe6df120988d84a9f6657c12573daR853

[wjwwood]

Yeah, that is the problem, it seems like there is an issue with the calls to PointCloudTransformer::transform and the new fallback behavior is being used.

If I disable the check of the res, and therefore make it so that the corresponding code block never gets called, the problem is fixed.

@vrabaud should we just remove the if (!res) case or should we fix it? Maybe we just need another check for arbitrary types, i.e. not just position and color? What is the purpose of that check?

[wjwwood]

Since this only affects visualization of intensity for lasers, and presumably non position or color fields in point clouds I am going to wait to iterate with @vrabaud rather than reverting the commit and releasing immediately.

Does that sound ok @jihoonl @mikeferguson?

[mikeferguson]

That strategy is fine for my use cases (I don't use intensity) but obviously can't speak for everyone.

[wjwwood]

Well I should also point out that this is not in public debians for hydro yet either, indigo it is, but I imagine that affects less people at the moment.

[jihoonl]

As long as it does not go into public, I am good too!

[vrabaud]

ok, I made a PR to fix it: the original intent was to only do a resize without a default value if possible (I believe it is faster).

[wjwwood]

I am going to backport #764 to hydro and then could you guys check to make sure that it addresses you problem? I am confident it will though because I tested it locally.

Thanks again to you guys for testing it all out, and thanks to @vrabaud for fixing it up.

[wjwwood]

hydro back port: #766