mutable iteration over mutable type produces a codegen error

[disruptek]

This is a portion of #12945 that wasn't solved.

iterator mpairs*[K, V](group: var Group[K, V]): tuple[key: K, val: var V] =
  for key, value in group.table.mpairs:
    yield (key, value)

This syntax produces a codegen error:

/home/adavidoff/.cache/nim/nimph_r/@[email protected]: In function ‘addedRequirements__NqsTCaB70tMWsL58CQq9czQ’:
/home/adavidoff/.cache/nim/nimph_r/@[email protected]:3671:24: error: ‘*dependencies’ is a pointer; did you mean to use ‘->’?
     T4_ = &dependencies->Sup;
                        ^~
                        ->

Tested in 1.0.6 and devel.

[krux02]

This is not a reproducible example.

[ghost]

@disruptek Can't reproduce on latest devel:

import tables
type
  Group[K, V] = object
    table: Table[K, V]

iterator mpairs*[K, V](group: var Group[K, V]): tuple[key: K, val: var V] =
  for key, value in group.table.mpairs:
    yield (key, value)

var grp = Group[string, string](table: initTable[string, string]())
grp.table["hello"] = "world"
grp.table["yes"] = "no"

for (k, v) in grp.mpairs():
  v = "hi"

[disruptek]

Here's my repro, which still demonstrates the problem:

https://github.com/disruptek/nimph/blob/master/src/nimph/group.nim#L131

[krux02]

@disruptek the comment I made about not being reproducable means that the issue should contain a code listing according to the issue template that can be compiled without any further change to reproduce the problem.

[disruptek]

It's a nice idea but I'm loathe to release a broken version of nimph just to satisfy the issue template. The fact that I'm currently unable to reduce the bug doesn't make the bug any less real.

[krux02]

Well then you should at least mention in the issue that you were unable to reduce the bug to a small example. The problem here is, back when I worked on bugs I skipped issues that don't bother to put in a reproducable example. And how should I distinguish from "you tried really hard to reduce the problem but couldn't" from "You didn't bother to put in a reproducable example", if you don't explain why you couln't reduce the problem.

[disruptek]

If I knew why I couldn’t reduce the bug then we wouldn’t be having this conversation.

If you want to help without working on this vague bug, I’m working on a tool to reduce bugs. It’s very crude and has grown less ambitious as I’ve run up against problems in the compiler, but it’s arguably a better place to type than GitHub comments:

https://github.com/disruptek/dust

It doesn’t yet attempt to reduce backend bugs such as this one. Maybe that would make a good first PR?