AssertEq seems to reject any operand other than invoke or const

[kg]

I can't figure out why it does this, but asserteq produces a syntax error if its operands are anything other than invoke or const. loads/loadu definitely don't work.

[rossberg]

That's working as intended, and is hard-coded in the syntax, see the grammar. The only way to access a module externally is by invoking an export, so you shouldn't be able to (or need to) observe anything else in a test.

[kg]

That's working as intended, and is hard-coded in the syntax

| AssertEq of string * Ast.expr list * Ast.expr list?

[rossberg]

Right, that's the (script) AST. The parsing rule is:

  | LPAR ASSERTEQ LPAR INVOKE TEXT expr_list RPAR expr_list RPAR
    { AssertEq ($5, $6 (c0 ()), $8 (c0 ())) @@ at() }

Those expressions are evaluated in an empty dummy context, so you (intentionally) don't have access to any state. IOW, constant expressions only -- this is intended as a declarative assertion about the behaviour of exported functions.

[kg]

OK, I see. So for the purposes of the ml-proto, we're acting as if each module has its own isolated heap (not the unified heap we will have post-dynamic-linking) and it's not exposed externally (unlike what will happen in embedded environments). And to do heap assertions you shove an assertion routine into the module? Awkward but reasonable, I guess. I don't like having to shove test code inside the module being tested :(

[rossberg]

Yes. @titzer was particularly adamant about not having a richer assertion
language, see the discussion on PR #14.

On 29 August 2015 at 04:05, Katelyn Gadd [email protected] wrote:

OK, I see. So for the purposes of the ml-proto, we're acting as if each
module has its own isolated heap (not the unified heap we will have
post-dynamic-linking) and it's not exposed externally (unlike what will
happen in embedded environments). And to do heap assertions you shove an
assertion routine into the module? Awkward but reasonable, I guess. I don't
like having to shove test code inside the module being tested :(

—
Reply to this email directly or view it on GitHub
#34 (comment).

[lukewagner]

@kg Even with dynamic linking, there will not be a single "unified heap" but, rather, explicit linking of new code into existing module instances. In ml-proto, the static-code/loaded-instance distinction is reflected by the Ast.modul / Eval.module_instance types. When we add dynamic linking, the logical extension would be to change Eval.module_instance to contain a list of Ast.modul. (To prepare for dynamic linking and to avoid confusion like in #306, I've been thinking we should rename Eval.module_instance to something that doesn't imply 1:1. Probably should update design repo first, though.)