Fix race condition

[marcoesposito1988]

As soon as a subscriber is started, it is able to receive messages and call the registered callback. However unlikely, this could cause the callback to access parameters and publishers which have not been inited yet. This race condition can cause obscure segmentation faults.

[bmagyar]

Hi there,

Please remove the commit "Ignore KDevelop files" from this PR.

The dynamic reconfigure removal is a good catch!

Has this race condition you mention ever appeared for you? Nothing should happen with callbacks until ros::spinOnce() is called later in the code, hence I find it unlikely to get into a race condition.

[marcoesposito1988]

Hi @bmagyar,

ok, I will remove the commit.

Not yet on this version of the code, but I am also trying to update the ArUco package and I got a segmentation fault. So I started poking around in the code and this jumped to my eyes. Even if this might not cause errors in this case, it is less error-prone and more future-proof to perform the initialization in this order (param reading -> publisher creation -> subscriber registration). Later modification to the code might introduce dependencies of the publishers from the params for example, and that would induce very subtle errors. So I thought of opening this PR.

[bmagyar]

Why do you need to update the aruco library? I thought of it before too but frankly it just didn't seem worth to go into that.

ROS Kinetic has OpenCV3 that comes with Aruco inside, the new version of this package will be pretty minimal. I intend to simplify the nodes too, probably only keeping one that's configurable.

[marcoesposito1988]

Oh I didn't know about the OpenCV integration! Yes, that is a good point.

I wanted to give it a shot because I also incur in issue #20. Markers are not detected as soon as they are further away than a certain distance from the camera. This even if the image is very good.

[marcoesposito1988]

BTW: I got the new version working! And it indeed works for longer distances, it appears.

Interested in a new PR?

P.S.: works, but not yet correctly.

[Flacedoo]

Is there a solution yet for the problem that markers are not recognized from a further than a specific distance? I ran a simulation, and when the camera is further away than 1m from a 6.5cm tag, it hasn't got any chance to detect it. Other libraries like apriltag has no problem at all with more than double the distance!

[bmagyar]

Hi guys, Yes, please do send in a new version on a PR against `kinetic-devel`. I've just created the branch!

…

On 2 February 2017 at 11:14, Flacedoo ***@***.***> wrote: Is there a solution yet for the problem that markers are not recognized from a further than a specific distance? I ran a simulation, and when the camera is further away than 1m from a 6.5cm tag, it hasn't got any chance to detect it. Other libraries like apriltag has no problem at all with more than double the distance! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#22 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADXH4X-WoCwrYsIeR-aiDN_5lgDFH4o7ks5rYayBgaJpZM4J6hdH> .