What?

This PR removesPrefixContext, and replaces it with an equivalent field on theModel struct.

Really?

I actually quite dislike this PR, and I don't think this is the correct direction to go in.

I thinkPrefixContext was easier to reason about because it allowed for a cleaner isolation of prefixing-related code. The absence of PrefixContext means that prefixes now get handled at different levels of the codebase, and is particularly messy when it comes to submodels.

But, well, the work's done, so I may as well put it up.

More details?

There are two main 'types' of changes in this PR:

(1) Compiler

Inside the compiler, instead of justvn (which is the lhs of tilde) we now have to handle bothleft_vn (which is the lhs of tilde) as well asvn (which is the result of applying any currently active prefixes).

In effect, because the prefixing no longer happens in the tilde-pipeline, prefixes have to be handled at a higher level. For example, previously you could do something like:

f(ctx::PrefixContext, vn::VarName)=f(childcontext(ctx),prefix(ctx.prefix, vn))f(ctx::OtherContext, vn::VarName)=...

SincePrefixContext no longer exists, you have to prefix the VarName in the compiler itself before passing that to the general methodf(ctx, vn).

This strategy works for functions likegetconditioned,hasconditioned, etc.

(2) tilde-pipeline

Fortilde_assume!!, the above strategy is not enough. We also have to threadmodel.prefix through totilde_assume!!.

The reason for this is quite subtle and has to do with submodels. Consider the following:

@modelfunctionf()    x~Normal()# (2)end@modelfunctiong()return a~to_submodel(f())# (1)endpg=prefix(g(),@varname(hello))

In this case, we want the sampled variable to be calledhello.a.x. If we use automatic prefixing as is shown here, this is easy to do: insidetilde_assume!! of thetop-level model (1),a will already have been prefixed withhello, so theleft argument totilde_assume!! will behello.a. We can thus simply set this as the appropriate 'prefix' for the inner model.

The problem arises if we don't automatically prefix:

@modelfunctionf()    x~Normal()# (2)end@modelfunctiong()return a~to_submodel(f(),false)# (1)endpg=prefix(g(),@varname(hello))

If we run this, everything is the same as before: when we get to (1),tilde_assume!! knows that the prefixed left-hand side ishello.a.However, because we don't want to prefix the submodel, the inner variable should only be calledhello.x, andnothello.a.x.

This means that we have to store the prefixhello separately so that we can apply it to the submodelwithout thea as well.

Note that in the absence of submodels, this would be entirely unnecessary.