AUv2: initialize the Cocoa view before creating the plugin's GUI

[defiantnerd]

Fixes #582. An alternative to #591 — thanks @busticated for the report, the stack trace and the patch; this takes the first option you suggested in the issue.

What happens today

initWithAUv2: registers self as the CLAP window and runs gui->create() before [super initWithFrame:]. A plugin that calls request_resize from inside create() — which clap/ext/gui.h allows — therefore reaches -setFrame: on a view AppKit has not finished building, and NSView throws NSInternalInconsistencyException: '_trackingAreaHelper not allocated yet!'.

What this does

Initializes the view first, with the size the host offered, and creates the plugin's GUI into a view that is ready for it. A request_resize during create() then lands on the frame like any other resize, and the frame it leaves behind becomes the initial size; get_size() is only asked when create() did not ask for a size of its own, so the plugin's own request still wins.

canary moves to after _createWindow(). It says the view is connected to a live plugin GUI — which is what every other reader of it means (viewDidMoveToWindow, dealloc, -setFrame:, and _destroyWindow in wrapasauv2.cpp) — and it should not say so while create() is still running: the set_size() echo that -setFrame: normally hands back to the plugin would re-enter create(). That is the same reentrancy #591 avoids with its early return.

So no new state on the view: "may -setFrame: talk to the plugin yet" is canary, and "did create() ask for a size" is the frame having moved. Registering the window after [super initWithFrame:] also fixes a latent wrinkle — self and &canary were handed to the wrapper from an object super init had not returned yet.

Open question

Does the requested size really need to beat get_size()? After a request_resize in create(), a conforming plugin's get_size() should return the new size, in which case the NSEqualSizes branch can go and the fix is purely the reordering. @busticated — does nice-plug-egui's get_size() disagree with what it asked for? If it does, that reason should be written into the comment.

Testing

Builds all three AUv2 targets. Not yet verified in a host — auval cannot open an editor, so this wants the pluginval editor test from #582 and a look in Logic, where the host follows the view's frame-change notification.

[busticated]

ah, yeah that's much nicer - thanks 🙏👍

confirmed this fixes #582 for me.

here's how i repro'ed things: a nice-plug@0.4.2 / nice-plug-egui@0.5.1 plugin built through clap-wrapper-rs (repo), pluginval@1.0.4 (strictness 5 | repo).

• stock v0.16.0 aborts in the AU editor test right after creating NSView
• v0.16.0 with this commit passes clap-validator@0.4.1, pluginval (VST3 and AU) and auval.

NOTE: I applied the commit to v0.16.0 by hand, leaving out the #536 entry guard, which isn't in that release.

i'll close #591 in favor of this. thanks again for working on this stuff.