John Yani

The suspend/resume detection is not a solution. There are cases where device is lost without suspension. And there are cases where suspension does not cause the device loss. The only...

The context of `wait_for` being stuck for seconds and the broader design around handling GPU device loss in suspend/resume scenario is described in detail in [this issue](https://github.com/zed-industries/zed/issues/23288). Please comment if...

Rereading this, I think now I understand better. You're implying that wait_for is not the only place where Zed might be stuck when unrecoverable error occurs? We're still discussing this...

> Hardware errors: It's acceptable to crash the application. Let the application decide please > Execution timeouts: These should be treated as implementation bugs and fixed. Huh? You're saying it's...

So, would you willing to review a PR that surfaces some kind of error to the application? ### Step 1: Update the `CommandDevice` trait Add the new method `wait_for_with_error` Something...

@kvark Ok. Some further reflection on the issue I'm facing. The issue appears to happen when prime-select is in performance mode. So I think it means that in this case...

> Not sure I understand this part. How can we be picking an adapter if it's turned off? We are doing all the healthy Vulkan checks on a VkPhysicalDevice before...

Interesting results in on demand mode: https://github.com/Vanuan/zed_nvidia_prime/blob/189e353af37bf4a2e6ecb8450c5d47692cb6d796/test_results/prime_select_on_demand/zed_nvidia_icd/logs.txt#L24-L47 This confirms that `vk::Result::ERROR_DEVICE_LOST` error occurs during suspend (at least in demand mode): https://github.com/kvark/blade/blob/4588302a1fc53ef1b2794a390b513084db4b8d45/blade-graphics/src/vulkan/command.rs#L430 On the other hand, in performance mode, this issue...

Could be a duplicate of this: - https://github.com/zed-industries/zed/issues/7940 I've investigated the issue for a potential solution. PTAL: - https://github.com/zed-industries/zed/issues/23288 Though this is not solving root cause, it could provide an...