Martin Berger

@allenjbaum We do not need enormous string of command line options. We can ask the C simulator to take an additional command line argument, which might take a file name...

Luckily Sail can define option types [1] and they are widely used in the RISCV model [2]. It looks like the type you are sketching is a list of 4-tuples,...

_> define new types at run time_ We cannot define Sail types at (Sail or C) run-time.

> IF we could tag it with "wishlist", Do we have a place where we put wishes and the like?

> I'm not asking for all of these, but ... Good idea. This is supported by Github anyway.,

The PMA spec leaves open the question whether PMA ranges can overlap or not. If overlapping is allowed, a host of issues arise, including: - Do do ranges need to...

@allenjbaum How about something like this: For every register: (CSRs and others) - An explicit Sail write-legalisation function, that defaults to the identity (i.e. every value is legal), except where...

Agreed. As Sail is more widely used, and the ISA community is becoming more familiar with the language, we should move the model towards more type safety and less repetition.

I'm also in favour of extending the `Retired` data type. There is another important case: Sdtrig debug triggers (https://github.com/riscv/riscv-debug-spec). One way to implement is to centralise trigger handling uniformly in...

There is a minor issue here in that Sail's type abbreviations (like `type MemoryOpResult('a : Type) = ...`) are *structural* rather than *nominal*. This might be most easy to see...