Add support for custom printing and parsing

[math-fehr]

Should fix #113

This PR allows operations to override print and parse to essentially implement custom parsing. It also open a bit the API of the printer and parser class.
This PR does not add custom format for dialects defined in xDSL, so there should not be any breaking change.

There is an example of its usage in tests/printer_test.py

[webmiche]

Looks very good for me 👍

What would be even cooler is if we could declaratively describe what the printed op should look and it would generate parser and printer for free. Like the PlusCostumFormatOp is specificied as:

"test.add" "lhs" + "rhs"

[webmiche]

How long did you spend on those? xD

[Dinistro]

I hope that he used a tool for that ;P

[math-fehr]

Don't worry, I use figlet ;)

[Dinistro]

At a first glance this seems to look good. Unfortunately, I don't have time today for an in depth review. Not sure if that would be required though.

[webmiche]

In terms of op definition generations, @nicolasvasilache pointed me today to OpDSL(Git), a DSL that allows defining operations from which linalg.generic is produced. Might be interesting to understand how this correlates with what we are trying to do :)

[webmiche]

Food for thought: It might be cool, if we could just write something like

printer.print(f"{self.lhs} + {self.rhs}")

here, tough I guess that does not work in our current infrastructure as the SSAValue names are not known...

[math-fehr]

Yeah, the problem here is that we cannot just do a print(self.lhs), since we need the context.
I'll simplify it with printer.print(self.lhs, " + ", self.rhs) if you want!

[webmiche]

Yes that would look better I feel. You mentioned ssomething about refactoring printer and parser at some point, maybe we can make it such that this would be possible afterward?

[math-fehr]

Yeah, though I'm not completely sure that I'll be able to do that :/
The problem is that I don't know how we can print an SSA Value without a context

[math-fehr]

Though what we could do is to write assembly_format = "$lhs + $rhs" if you want.
Though we could do this in a separate PR

[webmiche]

Yeah let's leave this one as is, I think its fine like this.

[math-fehr]

In terms of op definition generations, @nicolasvasilache pointed me today to OpDSL(Git), a DSL that allows defining operations from which linalg.generic is produced. Might be interesting to understand how this correlates with what we are trying to do :)

Yes, OpDSL is kind of related to IRDL as well, though OpDSL is specialized for linalg.
Compared to what we are doing, this OpDSL would essentially be a front-end DSL that targets IRDL.

[math-fehr]

I simplified things a bit, now it looks like printer.print(" ", self.lhs, " + ", self.rhs).
Of course, we would like to have something like assembly_format = "$lhs + $rhs" at some point, but we should do it in a separate PR I feel, since this can be a lot of work.

[math-fehr]

If this is good for you, I think we can merge it!
This shouldn't have any breaking change (since I don't think we were using the private functions of the parser/printer).