IIUC, the only reason gopls makes workspace/configuration requests is to support different settings for different workspace folders. In particular, this allows for different sets of build tags per folder, which is a workaround for lacking build tag support in gopls (#29202). I'm not sure whether anyone is actually using this workaround.

In VS Code, per-folder setting is an essential part of handling multi-root workspace environment (https://code.visualstudio.com/docs/editor/multi-root-workspaces#_settings). I am afraid dropping per-root folder configuration without an alternative solution will result in more confusion for most VS Code users.

Multi-root workspace feature allows users to create a workspace the users interact with day-to-day in one window. For example, a user can have her small size personal project that depends on a k8s module and the k8s project in one VS Code workspace. In this setting a user may want to enable more expensive analyzers for her own project, but that doesn't necessarily mean she also wants to enable them for all projects in the VS Code workspace. Similarly, a user may want to toggle the new vulncheck scanning feature in one folder, but doesn't want to enable it in all projects in the workspace. Without workspace/configuration, implementing this behavior will be challenging (unless we choose to start a separate gopls instance per each registered folder, which I don't think is great).

My interpretation of microsoft/language-server-protocol#972 is that it is about the protocol design, not about dropping the per-folder configuration feature that relies on LSP workspace/configuration.

As mentioned in #54559 (comment) this functionality is controlled by the workspace.configuration capability, and editors that cannot handle this properly shouldn't advertise the capability.

Re: #54559 (comment)

The problem is that the documentSymbol result in (3) depends on the configuration returned in (2); we don't produce semantic tokens without packages, and we don't load packages without a Go environment, and we don't have a Go environment without configuration. Therefore, if the configuration request in (2) is blocked, gopls has the choice of either serving a degraded result in (3), or timing out.

I think it's reasonable for language servers to block those requests until the configuration is ready, unless the client indicates the requests are server cancellable (LSP 3.17 severCancelSupport - from microsoft/vscode#161627 (comment) I guess VS supports that only for semantic tokens). Clients will need to cancel the request after some time or wait patiently depending on their need.