Nim tries to resolve generic field type name even though when condition is false

[endragor]

Sample:

const hasThreadSupport = compileOption("threads")

type
  SomeObj*[T] = object
    when hasThreadSupport:
      channel: ptr Channel[T]
      # Error: undeclared identifier: 'Channel'

The error occurs when you try to compile this without --threads:on option.
That is only specific to generic types.

Nim Git hash: 9784e36

[dom96]

Hrm, could this be the same as #4797?

[endragor]

Yeah, looks the same.

[nc-x]

The reason is that in generic contexts, we do not set the correct branch to be evaluated and instead evaluate all the branches

Nim/compiler/semtypes.nim

Lines 663 to 668 in ab72d68

	if c.inGenericContext == 0:
	var e = semConstBoolExpr(c, it.sons[0])
	if e.kind != nkIntLit: internalError(c.config, e.info, "semRecordNodeAux")
	elif e.intVal != 0 and branch == nil: branch = it.sons[1]
	else:
	it.sons[0] = forceBool(c, semExprWithType(c, it.sons[0]))

It can be fixed currently for simple boolean expressions I think.
But it cannot be fixed currently for expressions like when T is SomeType.
Because semIs does not evaluate the expression at that moment so we cannot know which branch to take

Nim/compiler/semexprs.nim

Lines 407 to 409 in ab72d68

	if c.inGenericContext > 0 and lhsType.base.containsGenericType:
	# BUGFIX: don't evaluate this too early: ``T is void``
	return

IDK what can be done here.
The only possibility I think is to move the check to instantiation time for generic objects.

[Araq]

This is really nasty to fix...

[metagn]

Was fixed in #24042