const cstring incorrectly cgen'd (pointer is copied, pointing to garbage)

[yglukhov]

type O = object
  a: cstring

proc test(o: var O) =
  const foo = "Hello"
  const fooc: cstring = foo
  o.a = fooc

var o: O
test(o)
echo o.a

Actual output: empty
Expected output:

Hello

Offending commit: 85db42a
The issue can be worked around with the following pattern: o.a = static(cstring(foo))

[Clyybber]

I think this may be related to #12282 (EDIT(timotheecour): probably not, this doesn't involve distinct), since with 85db42a there should now be an implicit conversion in const fooc: cstring = foo.

[timotheecour]

reduced:

when true: # D20201024T083352
  const a = "foo"
  const b: cstring = a
  # const b: cstring = "foo" # would work
  var c = b
  echo (a.repr, b.repr, c.repr, a, b, $b, c, c.len, cast[int](c), cast[int](b))

prints:
("0x105df7f98"foo"", "0x105df7f98"foo"", "0x105df7f88"\3"\n", "foo", "\x03", "foo", "\x03", 1, 4393500552, 4393500552)

so the title is wrong, the cstring is not empty, it's just that the cstring is being copied incorrectly (the pointer is copied directly instead of copying the content it points to) when crossing the VM => RT boundary

note

another unrelated issue I found is:

when true: # D20201024T085220:here
  const a = "foo"
  var b: cstring = a
  echo (a.repr, b.repr)

("0x10d7dff98"foo"", "0x10d7df518"foo"\n")
=> repr gives different results for VM cstring vs RT cstring; this is likely due to how cstring's are represented in VM so might not be a real bug, but it's still an inconsistency