data drop / fifo overrun

[jdavidberger]

Are there any plans to offer any thing like just latching in the last value on the ports, or somewhat equivalently, a FIFO that replaces the oldest value when it overflows?

Would be useful when you are processing data live. The specific use case I'm looking at is using it to prototype opencv filters.

[jonathan-beard]

There are actually. This is similar to the leaky fifo issue. There is an implementation of the fifo interface already there tray will do most of what we want. It was created a long time ago for testing. Needs to be dusted off and to have hooks added to the allocator factories. Will push that higher in the work queue unless you would like to volunteer? I'm happy to hand out member keys to commit to the project.

…

-Jonathan Sent from my tiny screen, powered by ARM awesomeness.

On Mar 12, 2017, at 05:02, jdavidberger ***@***.***> wrote: Are there any plans to offer any thing like just latching in the last value on the ports, or somewhat equivalently, a FIFO that replaces the oldest value when it overflows? Would be useful when you are processing data live. The specific use case I'm looking at is using it to prototype opencv filters. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[jdavidberger]

I looked into some of the code that was there and wasn't sure what the intention was for some of that.

Are you thinking this should be a run-time parameter you can set on the FIFO -- something like setOverflowBehavior -- or as a separate implementation type entirely put in the const_map for port_info?

[jonathan-beard]

If you check out the dev branch, the intention would be to set a stream modifier for the indented behavior for that link. I'm rebuilding the template parser at the moment. The code is largely there. Need to fill in the templates. Basically I've ended up making a DSL using the templates to express graphs. I realized the previous parser had a few extensibility limitations so I spent some time making it easier to read and use. For the behavior you describe raft::manip would likely be the perfect place.

…

-Jonathan Sent from my tiny screen, powered by ARM awesomeness.

On Mar 12, 2017, at 16:20, jdavidberger ***@***.***> wrote: I looked into some of the code that was there and wasn't sure what the intention was for some of that. Are you thinking this should be a run-time parameter you can set on the FIFO -- something like setOverflowBehavior -- or as a separate implementation type entirely put in the const_map for port_info? — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[jdavidberger]

I checked out dev; seems like a lot is ifdef'ed out for the refactor so I couldn't trace through it while running.

It seems like overrun behavior should be settable on the port and not the kernel. So would the syntax look something like:

M += s["0"] >> raft::manip< raft::fifo::drop_overrun >::value >> l;

to set the fifo to drop overruns for s["0"]?

That seems to imply that this is a run-time settable attribute on the FIFO -- by the time any of the stream operators kick in, the ports are already made -- so would a PR where you can set the behavior directly on the FIFO be helpful until the refactor with more manip functionality is done?

[jonathan-beard]

This one will be added into the improved parser over my current vacation :).

I'd rather not create yet another syntax option that would fragment the environment. The only problem with the runtime settable version is creating more than one way of doing things...which isn't a problem in and of itself, but does make it harder for people to learn.