Use of delayed and physical connections as provided by the C++ runtime #1583
Conversation
This reverts commit 1bd0fa4.
|
@petervdonovan Could you take a look at the failing LSP tests? If you have any tips on where I can start looking myself in the future (I already struggle with finding the actual error in the log), I would appreciate that. Or even better: it would be great if we could have some instructions on how to handle the LSP tests in the developer docs. |
|
The LSP tests appear to be passing
I will try to do this tomorrow or the day after tomorrow |
Oh, then I guess it was just one of the regular C++ tests also causing the LSP test failures.
That would be great! Thanks! |
91ab496 to
4cb0758
Compare
org.lflang/src/org/lflang/generator/cpp/CppConnectionGenerator.kt
Outdated
Show resolved
Hide resolved
| import org.lflang.lf.Reactor | ||
| import org.lflang.lf.VarRef | ||
|
|
||
| class CppConnectionGenerator(private val reactor: Reactor) { |
There was a problem hiding this comment.
This class needs some documentation...
Currently, after delays are implemented as an AST transformation. With lf-lang/reactor-cpp#38, the C++ runtime provides a native solution for implementing delayed and physical connections. This PR disables the AST transformation in the C++ target and instead uses the native runtime implementations. It also adds a couple of tests for various corner cases that have not been tested before.
In short, the runtime provides to classes
DelayedConnectionandPhysicalConnectionthat are specializations of theActionclass (which better should be called Trigger). By implementingAction, the connections may add new events to the event queue. At the same time, they provide API for binding to one upstream and multiple downstream ports. Due to a callback mechanism, the connection gets notified when the upstream port is set and it will schedule an event. When the scheduler retrieves the event from the event queue, itssetup()function will also set all downstream ports and accordingly trigger downstream reactions.This PR probably makes the C++ target the first target to correctly implement physical connections, as physical connections are currently not properly handled by the AST transformation (see #616)