V3Scope: Check upper scopes when searching for scope in VarRef

[kamilrakoczy]

As described in 23.9 Scope rules of IEEE, if identifier isn't declared in local scope, we should search scopes upward until we find valid identifier.

[wsnyder]

All searching is done in V3LinkDot, the V3Scope should only be about flattening with all of the var pointers otherwise already correct. I suspect the input .tree to V3Scope was wrong, please check, or I can look tonight.

[kamilrakoczy]

I've checked input .tree, but I think it looks ok, V3LinkDot correctly linked this VarRef to top level variable:

    1: MODULE 0x555556472fc0 <e1098> {f7ai}  t  L2 [1ps]
[...]
     1:2: VAR 0x555556473280 <e1040> {f8aj} @dt=0x5555564731b0@(G/w1)  read_data [VSTATIC]  VAR
[...]
     1:2: CLASS 0x555556473950 <e639> {f12ad}  monitor_concrete  L3 [1ps]
[...]
     1:2:2:3:1: VARREF 0x555556455470 <e1066> {f14ay} @dt=0x5555564731b0@(G/w1)  read_data [RV] <- VAR 0x555556473280 <e1040> {f8aj} @dt=0x5555564731b0@(G/w1)  read_data [VSTATIC]  VAR
[...]

I think the problem is when we are visiting VarRef in V3Scope:

verilator/src/V3Scope.cpp

Line 291 in 4990b44

m_varRefScopes.emplace(nodep, m_scopep);

we are adding here current scope (in relation to VarRef, not Var). I actually tried to get AstVarScope (and AstScope in the end) by using user1p of nodep->varp(), but for some reason I was getting nullptr (I suspect, that we might first visit VarRef and then Var, but I haven't investigated this fully).

[wsnyder]

we are adding here current scope (in relation to VarRef, not Var).

Well that's definitely wrong, are you sure? Think lots would have broken before now.

varp() should never be null. It might be though that the var itself is in a scope that will be visited later than the varref. So we'd need to record the information in a varref->var map, and another var->scope map, then use both maps to varref->var->scope in a final loop in the end.