Fixed incomplete type declaration

[BigBoyBarney]

This fixes the issue I opened in the GTK4.cr repo.
I have no idea if this breaks anything else.
FWIW the specs with make test passed 🤷

[hugopl]

If it's passing on CI os a good sign, but would be nice to add a test for the use case it fixes. I'm without access to a computer for few days, but can help you with this commenting from my phone.

The ideia is to create a method in libtest that mimic the buggy method found in GTK4, then add a test that call it and check if everything was ok.

[BigBoyBarney]

Libtest contains C code. Normally I would write a spec like

it "correctly splices an array of objects into a ListModel" do
  list_store = Gio::ListStore.new(Gtk::Button.g_type)
  one = Gtk::Button.new
  items = [one]
  list_store.splice(position: 0_u32, n_removals: 0_u32, additions: items)
end

But I suppose that's not what you had in mind :D

[hugopl]

If you can reproduce the issue with modules included in gi-crystal is even better!

Sometimes it's not possible because the use case is found just in GTK, so a libtest code is needed since gi-crystal can't depend on GTK.

[BigBoyBarney]

Ahhh, okay, alright then!
The issue was an argument type mismatch, so it was a compile-time error in Crystal.

Sorry for the silly questions, but I can't write the aforementioned spec, because it can't find things like Gtk::Button. 🤣, even though it's in the generated bindings auto/gtk4. What's the intended way of doing this?

Also, since the above example is an argument type mismatch at compile time, the spec wouldn't even be able to "run" if it failed. Not sure if that's a good thing or not 🤣

[hugopl]

So it's the case for a libtest implementation. I'm moving right now, without Internet and my office is inside boxes 😅, if I find some time I will check the issue in detail and can tell the directions to reproduce the issue in libtest.

[hugopl]

You can use Test::Subject instead of Gtk::Button and check if it also fail.

[BigBoyBarney]

The following spec

describe "ListStore" do
  it "can splice an array" do
    list_store = Gio::ListStore.new(Test::Subject.g_type)
    one = Test::Subject.new
    items = [one]
    list_store.splice(position: 0_u32, n_removals: 0_u32, additions: items)
  end
end

fails on current main with the compile-time errors:

In spec/list_spec.cr:65:18

 65 | list_store.splice(position: 0_u32, n_removals: 0_u32, additions: items)
                 ^-----
Error: instantiating 'Gio::ListStore#splice(position: UInt32, n_removals: UInt32, additions: Array(Test::Subject))'


In src/auto/gio-2.0/list_store.cr:339:67

 339 | LibGio.g_list_store_splice(to_unsafe, position, n_removals, additions, n_additions)
                                                                   ^--------
Error: argument 'additions' of 'LibGio#g_list_store_splice' must be Pointer(LibGObject::Object), not Pointer(Pointer(Void))
make: *** [Makefile:4: test] Error 1

And passes in the PR. So in that regard, it "works" as proof, but the spec run erroring out on the main branch (without the PR) at compile time and aborting instead of failing that specific test isn't great.

[hugopl]

Add the test in this same PR that has the fix, so we merge all together.

[BigBoyBarney]

It fails the CI test but passes locally 🤔

[hugopl]

It fail just when compiling with -Ddebugmemory, so it probably pass by some coincidence when not using this flag that just print to stdout when a object is garbage collected.

[BigBoyBarney]

I edited the bin/spec script to include -Ddebug_memory like

#!/bin/env sh

export BUILD_DIR="$(pwd)/.."
export BUILD_DIR=$(dirname $BUILD_DIR)

# Let the linker know where libtest is
export GI_TYPELIB_PATH="$BUILD_DIR/spec/build"
export LIBRARY_PATH="$BUILD_DIR/spec/build"
export LD_LIBRARY_PATH="$BUILD_DIR/spec/build"
# export G_DEBUG=fatal-warnings

crystal spec -d --error-trace -Ddebugmemory $@

and it still passes locally.

Is there something else I should enable to get the same test that fails on CI?