win-capture: Avoid NULL deref when capture not initialized

[confusionattack]

Description

Add a check to avoid a NULL dereference on backbuffer capture in the D3D10, D3D11, and D3D12 capture modules in case d3d10_data, d3d11_data, or d3d12_data has not yet been successfully configured. This is similar to NULL checks in other places in these modules, such as the d3d10_free, d3d11_free, and d3d12_free functions. This change is intended to fix a specific issue with D3D12 capture we are seeing the wild, but for completeness we added NULL checks to the D3D10 and D3D11 capture modules.

Motivation and Context

We were seeing an intermittent crash in graphics-hook when using OBS Studio to capture the Midnight Murder Club demo (https://store.steampowered.com/app/2698870/Midnight_Murder_Club/) on some laptops with multiple GPUs. We tracked it down to a NULL dereference when capturing the backbuffer in the D3D12 capture module.

How Has This Been Tested?

We were able to a user who could reliably replicate the crash on an Intel i7-11800H laptop running Windows 11 with an NVIDIA GeForce RTX 3060 Laptop GPU and a built-in Intel UHD GPU. The repro steps were as follows:

• Start the MMC demo and progress the main menu.
• Create two scenes in OBS Studio. The first scene is a static image. The second scene is the game window capture.
• Rapidly toggle between the two scenes until a crash occurs.

After applying implementing the NULL check fix and testing against a locally built OBS Studio, we were unable to reproduce any issue.

Types of changes

• Bug fix (non-breaking change which fixes an issue)

Checklist:

• My code has been run through clang-format.
• I have read the contributing document.
• My code is not on the master branch.
• The code has been tested.
• All commit messages are properly formatted and commits squashed where appropriate.
• I have included updates to all appropriate documentation.

[jpark37]

Removed myself as a reviewer since I'm not active enough to be holding things up.
Thanks for the PR. I suspect my preference would be to resolve the condition leading this implied invariant to break rather than sprinkle around null checks. Even if that fix means sprinkling a null check somewhere else. :P Something about our understand of how the code can flow is probably flawed here, so it would be good to try to understand how this is happening before making changes. Not asking you to do the digging, but someone should.
I'll stay out of this from here and let Lain handle it.

[confusionattack]

I take jpark37's points. This change does not really get at root causes. It's little challenging on my end because I don't have physical access to any of the machines where this crash is being reported. But with more effort, I can determine the root issue on the graphics-hook side. Or maybe I should just focus on modifying the game client to avoid putting graphics-hook in a position where it crashes and then this PR won't be needed at all? Lain-B, do you have any thoughts?

[RytoEX]

For reference, I believe #11430 goes the "sprinkling a null check somewhere else" route by checking data.handle before trying to call any of the _capture functions.

@confusionattack does that PR (already merged) resolve your observed crashes?