Overload resolution broken with subclassing + or

[planetis-m]

In the code below, overload for Option is not used instead the generic object overload takes its place.

Example

import options, streams

proc storeBin*(s: Stream; x: bool) =
  write(s, $x)

proc storeBin*[T](s: Stream; o: Option[T]) =
  let isSome = isSome(o)
  storeBin(s, isSome)
  if isSome:
    storeBin(s, get(o))

proc storeBin*[T: object|tuple](s: Stream; o: T) =
  write(s, "bug")

var s = newStringStream()
s.storeBin(none(bool))
#s.storeBin(some(true)) #same
echo s.data

Current Output

bug

Expected Output

false

Additional Information

• Splitting object|tuple type class, into separate procs, fixes it.
• Using a var parameter in storeBin proc, fixes it.
• Wrapping storeBin into another generic proc, fixes it.
• Issue Error: overloaded 'X' leads to ambiguous calls #12787 is related.

$ nim -v
Nim Compiler Version 1.5.1 [Linux: amd64]
Compiled at 2020-12-09
Copyright (c) 2006-2020 by Andreas Rumpf

git hash: 9ce2f87a0a02d85e943b2b181bcae66fbd644f11

[planetis-m]

Another test:

type
  Foo*[K] = object
    item: K

template empty*[K](t: typedesc[K]): K = high(t)
proc initFoo*[K](): Foo[K] =
  result = Foo[K](item: empty(K))

type
  Bar = distinct uint16

proc `==`(a, b: Bar): bool {.borrow.}
template empty*(t: typedesc[Bar]): Bar = Bar(0)

var x = initFoo[Bar]()
assert x.item == Bar(0)

[timotheecour]

always try to minimize please ;-)
If it's a compiler bug, ideally it should be minimzed to not rely on other imports (or at least only hand-crafted imports).

Here's a minimized repro without imports, and here's an updated diagnostic:

overload resolution wrongly favors tyOr when some argument matches via subclassing.

when true:
  type Stream = object of RootObj
  type Stream2 = object of Stream
  type Foo[T] = object
  proc storeBin[T](s: Stream; o: Foo[T]) = discard
  proc storeBin[T: float|object](s: Stream; o: T) = doAssert false
  # var s = Stream() # ok
  var s = Stream2() # bug
  s.storeBin(Foo[int]())

links

sigmatch.ninm: see isSubtype, subtypeMatches, isGeneric

[planetis-m]

Thanks! I tried my best but you helped to make it even smaller.

[planetis-m]

I don't know if you want me to submit different issues, but second test case is not related to subclassing as far as I can tell :)

[timotheecour]

ya, it may be a different bug then.

Maybe try to take a stab at a PR? See sigmatch.nim, it sounds not too complicated (YMMV)

[planetis-m]

sure will try tomorrow