Is it possible that GHC generates STG with invalid binding semantics for certain cases that the Cmm codegen fix or ignore?

This could explain my observations.

I've checked the Stg linter source (StgLint.hs ; GHC 8.2.2 and github master) and it does not check StgRhsClosure free var and binder list at all.

And the scope checker function (addInScopeVars) does not check for duplicates.

Any thoughts?

Cheers,

Csaba

On Sat, Nov 3, 2018 at 9:53 AM Csaba Hruska <csaba.hruska@gmail.com> wrote:

Hi,

Can StgRhsClosure's freevar list ([occ]) or argument list ([bndr]) contain duplicates?

Cheers,
Csaba

data GenStgRhs bndr occ
= StgRhsClosure
        CostCentreStack         -- CCS to be attached (default is CurrentCCS)
        StgBinderInfo           -- Info about how this binder is used (see below)
        [occ]                   -- non-global free vars; a list, rather than
                                -- a set, because order is important
        !UpdateFlag             -- ReEntrant | Updatable | SingleEntry
        [bndr]                  -- arguments; if empty, then not a function;
                                -- as above, order is important.
        (GenStgExpr bndr occ)   -- body