cgen or internal error when passing @[] to seq[string] or T (ditto with {} and set[uint8])

[timotheecour]

passing @[] to seq[string] or OtherType gives a cgen error.

I use the following pattern as a very convenient way to have a compile time Option[T] type, however it doesn't work with T = seq[string] when passing @[]

Example 1

when true: # gitissue D20201026T002034:here
  proc fn(a = 4, b: seq[string] or tuple[] = ()) =
    discard # eg: when b is tuple[]: ...
  fn(1)
  fn(1, @[""])
  var a: seq[string] = @[]
  fn(1, a)
  fn(1, seq[string](@[]))
  fn(1, @[]) # BUG: error: conflicting types for 'fn__d58I39cH9a6bcpi3QDPJ5dBA'

Current Output

@mt11189.nim.c:354:31: error: conflicting types for 'fn__d58I39cH9a6bcpi3QDPJ5dBA'

Expected Output

works

Example 2

likewise with {} for set:

when true: # gitissue D20201026T002034
  proc fn(a = 4, b: set[uint8] or tuple[] = ()) =
    discard # eg: when b is tuple[]: ...
  fn(1)
  fn(1, {1'u8})
  var a: set[uint8] = {}
  fn(1, a)
  fn(1, set[uint8]({}))
  fn(1, {}) # BUG: internal error: invalid kind for lastOrd(tyEmpty)

Additional Information

devel 1.5.1 2265955
see also #15722

[disruptek]

The codegen is using TGenericSeq for the prototype and seqEmpty for the definition. Feels like something I should fix. 😁

[metagn]

seq[empty] matches seq[string] or T because it matches seq[string], but compiler cannot figure out how to convert seq[empty] to seq[string] or T so it stays as seq[empty], causes codegen error (confirm with echo typeof(b)). Kind of like #1214