class library: fix possible sync problem in NodeProxy clear

[telephon]

MixedBundle allows to schedule a function when sending happens. This
is what was used to schedule into the future the clearing of the old
group. It is much better to just delegate this to the server. Because
group objects are registered by NodeProxy with the NodeWatcher, it is
also released correctly. ( NodeWatcher.newFrom(s).nodes).

We need to add a very small offset between the bundle timestamps, so their order is guaranteed on the server.

[telephon]

(ping)

[telephon]

(ping)

[jamshark70]

I did test this a while ago -- sorry I didn't report back. I'm afraid I don't remember the exact result. I suppose it was successful -- if I had seen a problem, I would have reported it at that time.

I just retested found that I'm not getting an error from clearing a NodeProxy -- so I think the fix is good.

BTW retesting was a bit harder than it needed to be. Unfortunately this branch is now incompatible with the current yaml-cpp submodule, so:

$ git checkout telephon-topic-fix-nodeproxy-clear-timing 
error: The following untracked working tree files would be overwritten by checkout:
	external_libraries/yaml-cpp/.clang-format
	external_libraries/yaml-cpp/.codedocs
	external_libraries/yaml-cpp/.gitignore
	external_libraries/yaml-cpp/.travis.yml
Please move or remove them before you can switch branches.
Aborting

So what might have been just a couple of minutes -- switch branches, test, switch back -- became "try a couple of ways to stash some files that I'm not sure will be restored later because they are untracked," switch branches, test, switch back, discover that the four files didn't come back, and hope it doesn't make a difference the next time I have to build.

[telephon]

@jamshark70 yes, I stumble over it all the time: git has no very good way to deal with added submodules.

In some other pull request, @bagong showed me a way to switch branches correctly in such a situation, but I can't find it anymore now.

[bagong]

Yea, failing long time memory is one of the worst problems in command line oriented work (almost as bad as in coding ;) ) I spend most time switching between Google, Stackoverflow and the command line 😄. I think in this case the trick to make it memorizable is to think of a submodules as individual units that can be switched on and off. You address a single submodule by its path. Switch it on with init and switch it off with deinit. And get help with git help submodule. So if you switch to a branch where submodules and folder might conflict, switch the submodule off first...
The problem that can co-occur in this context is that you might need a clean build after a switch. clean to be taken literally hear: to make clean or cmake build . --target clean before switching. That will remove precompiled stuff that still expects the old code (haha, vaguely..). Deleting the build folder is possible of course too, but you will loose your build configuration in that case, which could be anything between irrelevant and pretty destructive...

[telephon]

(can be merged IMHO)

[mossheim]

If someone can give me some test code, I can test it out.

[jamshark70]

s.boot;

Ndef(\x, { SinOsc.ar(440) });

Ndef(\x).clear;  // node not found happens here

[mossheim]

Ah, i didn't realize until i read through #2818 that this is a Linux issue. I don't see that error message on this branch or master on macOS Sierra. Someone on linux should test too.

[telephon]

If I understood this correctly, this is a potential issue on other systems as well?

[telephon]

it's just a little lot hard to reproduce

[gusano]

@telephon @jamshark70 this fixes the issue for me on linux (3.9dev, 4.10.6-1-ARCH)

[gusano]

@telephon A trailing whitespace! : )

[telephon]

:)

[telephon]

[OT] I'd like to have http://www.cplusplus.com/reference/cmath/nextafter/ so we can just push up the time by a minimal timestep, because OSC doesn't support order apart from time (or bundles)

[gusano]

Danke @telephon !

[telephon]

you're welcome !