Auto re-tries

[shikokuchuo]

An update in mirai: 6a72261 I have disabled the NNG-level auto-retry mechanism.

This doesn't break any of your tests, but it is a change in behaviour so just checking in here with you. I forget if you still make use of this, or actually if this makes things easier for you.

The previous behaviour was that (assuming dispatcher) if a task crashes that the 'mirai' will remain unresolved. This behaviour isn't too helpful / transparent to a user as they would assume by default that it's still ongoing. You would need to know to check status() and see if the particular server was offline.

In particular, this also doesn't help with a use case like mirai_map(), where code like the following would have appeared to hang before the change (fortunately it's interruptible).

library(mirai)
daemons(3)
m <- mirai_map(list(a = "1", b = "2", c = "q()"), function(x) eval(str2lang(x)))
m[]

The new behaviour explicitly returns an errorValue, which makes it much easier to identify crashes. What to do can then be handled by user code.

If you have a valid use case for the old behaviour, just let me know and it shouldn't be a problem to enable via an argument (I'll then delay switching the default so it is non-breaking for crew).

[shikokuchuo]

I've now re-introduced the option at dispatcher() which gains a 'retry' argument in bd7fd6c, in light of your comments at shikokuchuo/mirai#117 (comment).

This will default to TRUE for the next release, and I haven't fully decided if this will change in the future, but crew can consider setting this in the daemons() call to be insulated from potential changes.

[wlandau]

Thank for doing that. After the next mirai release, I will set a default of TRUE for crew and allow users to set it to FALSE for debugging purposes (c.f. #170).