Mike J Innes

Ah, this is one thing that `recur!` should probably do that it doesn't currently. Basically the rand call gets lowered to ```julia self(Core._apply_iterate, rand, (arg1,), args) ``` but `apply_iterate` should...

I'm not sure I fully understand the question, you might need to give a bit more detail. In general that kind of syntax means that `%7` is a function, which...

Yup – `Block` is the public interface to all block operations, whereas `BasicBlock` is just the internal representation. Perhaps we need to be clear in the docs that internal fields...

It would be great to understand whether this actually comes from IRTools or the package (e.g. by showing that a no-op dynamo errors like this). And if so, helpful if...

This seems like a reasonable approach; it won't yet work for branches because `ir[v]` doesn't exist for block arguments. You instead have to enumerate each branch to the block and...

No, "data flow analysis" is just a general approach to writing algorithms on IR like this; see for example https://en.wikipedia.org/wiki/Data-flow_analysis. The clearest example in IRTools is probably [this function](https://github.com/MikeInnes/IRTools.jl/blob/ec381e9bfcac041334d2fb312cdd99bb2fc8a600/src/passes/passes.jl#L142) which...

No particular plan on my end, but if you want to add it I don’t see why not.

#42 address the second part of this. The first part is a bit tricky because we currently want the generated function to have a single `args...` slot so that the...

This is possible, of course, but *really* fiddly. `@dynamo` exists because it's pretty much impossible to get this right first try (it always takes me multiple attempts as well). We...

We've been `just_construct_ssa`-free for a while and the latest release passes tests on 1.5. I also just removed a last vestige of typed IR in https://github.com/MikeInnes/IRTools.jl/commit/93fa60dd2085fc9646e774f84f5b027ee7c8b611. A new problem on...