Recently the standardization meetings have been, or are close to running out of issues to talk about. This means that we are approaching the next step of the standardization of WebGPU and WGSL where they get to Candidate Recommendation status (CR). We should collectively decide what are the criteria for asking this transition based on spec / implementation / test statuses.

What do you think of the following criteria? Anything missing?

• The spec is semantic-complete:
  • All non-editorial issues have been resolved for WebGPU and WGSL.
  • All editorial issues (in the issue tracker or inline in the spec) about the semantics have been resolved. Issues about rewording / restructuring the spec can still be outstanding.
• The spec is implementable:
  • To the best of our knowledge one implementation (Chromium most likely given the current statuses) implements all the spec correctly and reasonably efficiently. (we always find subtle issues when implementing new parts of the spec, this makes sure at least one "debugging" pass has been done on each part of the spec).
  • Other implementations agree, without necessarily having done all of it, that the spec can be reasonably implemented in their architecture.
• The spec is reasonably tested ():
  • Each area in which WebGPU is novel is covered by reasonably exhaustive tests. (includes validation, new mechanisms like GPUExternalTexture, operations with DOM, buffer mapping, uniformity analysis, etc).
  • Each area in which WebGPU defers to target APIs has at least some simple tests to check arguments are correctly passed through (sampler wrap modes, rasterization, some WGSL builtins and arithmetic operations, ...)
• Cross functional approvals (unless stalled after multiple reminders):
  • Internal cross-functional reviews needed by each implementer gave approval for WebGPU (examples: Security teams, API owners, ...)
  • Public cross-functional reviews gave approval (W3C TAG, security, PING, ...)