Bugfix: Ref assemblies contained .property definitions at the wrong type in generated IL

[T-Gro]

A inner check for generating properties in reference assemblies was checking existence of generated methods.
However, before the fix, methods were only generated AFTER the property itself.

This lead to strange shift of property definitions, putting them to the very last type of a file.

Besides adding tests dedicated for this scenario, some existing tests had assertions for the (broken) IL - after knowing that the issue is in assignment of property into the wrong class, it was very visible.

[vzarytovskii]

@T-Gro please retarget the fix to 17.4 branch.

[vzarytovskii]

Wondering if we can rely on existence of properties when generating getter and setter methods? In other words if existing order matters and if we can theoretically break codegen by changing it?

Edit: probably shouldn't matter in this stage, since those only generate tables and indices in them

[psfinaki]

Yeah so I guess the right way to go would be to get rid of a shared state, but since it's a quickfix - thanks for adding tests :)

[vzarytovskii]

Yeah so I guess the right way to go would be to get rid of a shared state, but since it's a quickfix - thanks for adding tests :)

It's not just "getting rid" of mutable state unfortunately, due to how IL emitting works. It's big and complex, and will likely involve a nearly full rewrite. Perhaps reflection.emit can be used for it instead. I know @TIHan was looking into it.

It would be nice, but it'll likely be a quite low priority compared to all other issues we have now.

[T-Gro]

Wondering if we can rely on existence of properties when generating getter and setter methods? In other words if existing order matters and if we can theoretically break codegen by changing it?

Edit: probably shouldn't matter in this stage, since those only generate tables and indices in them

I have tested the generated IL and it was the same, the order does not matter for the layout of the output.

[psfinaki]

It's not just "getting rid" of mutable state unfortunately, due to how IL emitting works. It's big and complex, and will likely involve a nearly full rewrite. Perhaps reflection.emit can be used for it instead. I know @TIHan was looking into it.

It would be nice, but it'll likely be a quite low priority compared to all other issues we have now.

I never said "just" :) Sure.

[buvinghausen]

Thank you @T-Gro @vzarytovskii and team