parsing sets and matrices in Oxide

[spritezs]

Right now we can't parse matrices and sets in Oxide. The current DecisionVariable object is not enough for representing them (need to store their length) so we can either update it, or create a new Object. I will bring this up on Wednesday but please feel free to comment here.

[niklasdewally]

Hi Tudor,

I will be working on representing matrices, and the framework which we will use to represent and rewrite them and other higher level types, such as sets.

@ozgurakgun and I have a loose plan of how we want to approach this, and I will share an RFC on this next week once I've tidied my ideas into a concrete proposal.

In short, we will be adding representation rules like in Conjure, and use these for matrices and SAT encodings as well as higher-level types. The former are Savile Row level constructs, which uses a different abstraction; however, we think that representation rules will also apply to these lower level concepts.

Oxide internally represents things differently than Conjure, and uses a very different programming style, so our big question for now is how do we port these concepts over in a way that makes sense in Rust.

[niklasdewally]

Would be good to discuss this on Wednesday, bearing in mind that there are still quite a lot of unknowns here.

[niklasdewally]

As matrix length and all that in essence is in the domain, you could add it as a new type of domain internally too?

That would also allow it to be used in lettings

[spritezs]

I will come up with a rough implementation for lettings and discuss it on Wed.
But for stuff like "matrix indexed by" list(Domain, ",", "[]") "of" Domain we need to (from what I ve seen) at least create a new type of variable which can hold the additional info.

[ozgurakgun]

I think it might make sense for you two to have a brief chat to coordinate?

For context, @spritezs is mainly interested in converting sets to matrices, but since matrices do not exist yet there is room for collaboration here.

[niklasdewally]

I think it might make sense for you two to have a brief chat to coordinate?

For context, @spritezs is mainly interested in converting sets to matrices, but since matrices do not exist yet there is room for collaboration here.

We should definitely coordinate on representation rules and all that then.

[ozgurakgun]

Though I think @spritezs is initially thinking about how to extend the rust data types and extend the parser to read these in, so that shouldn't depend on the rule stuff.

[niklasdewally]

Though I think @spritezs is initially thinking about how to extend the rust data types and extend the parser to read these in, so that shouldn't depend on the rule stuff.

What I meant was that I'm not sure how representation rules will affect the symbol table, and whether we would need to do things in a specific way so that all the info is in the right places?

[spritezs]

@niklasdewally isn't our first worry representing matrices in the AST? I think that is a good stepping stone. What do you think?

[niklasdewally]

@niklasdewally isn't our first worry representing matrices in the AST? I think that is a good stepping stone. What do you think?

That's as good a place as any to start :)

[niklasdewally]

As we are both working around the symbol table, it would be helpful to me if you could split this into several smaller PRs if possible (if you don't already do so) so that we can keep our branches in sync.

If I submit any symbol table changes this week, I'll @-you so that I don't break anything you are planning.

[spritezs]

sounds great. i haven't thought about PRs so far so feel free to make some. i ll add additional ones along the way when i bump into issues.

If I submit any symbol table changes this week, I'll @-you so that I don't break anything you are planning.

same same. sounds good

[spritezs]

@niklasdewally Oxide doesn't allow for 'letting' statements. I modified the 'parse_model.rs' file to allow for simple 'letting b be 3' statements.

Will show you tomorrow and push if it's up to standards. Will look into letting more complicated structures next. Cya tmr!

[niklasdewally]

@niklasdewally Oxide doesn't allow for 'letting' statements. I modified the 'parse_model.rs' file to allow for simple 'letting b be 3' statements.

Will show you tomorrow and push if it's up to standards. Will look into letting more complicated structures next. Cya tmr!

Your fork might not be up to date, I added lettings in #619 and #621

[spritezs]

ok i think i pushed to the main repo instead of my own branch
sorry i l fix it now

[niklasdewally]

In particular, domain lettings, added in #621, can contain matrix domain declarations.

[spritezs]

Before I start coding I wanted to make sure I understood what needed modifying.

In terms of 'searching' for a set, what I need to do is modify the Domain enum and add another possible value. What I ve thought of is: DomainSet(Vec), but I don't really know what a set can contain. Is it integers? If so can the set be DomainSet(Vec<Range>)?

[ozgurakgun]

A set domain contains another domain in it. Remember the Domain type in Conjure we looked at together, that might help.