RFC-0022 · Design Q&A
Versioned draft snapshot
RFCs, reference catalogs, and evidence reflect Spec 3b2e3e7a114fb826b1445d39a48756b8fff629cd. Counts and pending/status notes belong to that snapshot. Read the dated project update for later progress.
Source repositories are private during staging. Links marked “repository access required” need access to the linked repository.
The combined trigger, compensation, data-flow, and retry proposal is in WORKFLOW_CONTRACT_PROPOSAL.md. The answers remain open until the authored RFC, generated catalog, and implementation agree.
Q: Which workflow trigger forms are valid?
Status: Open
Source: RFC-0022 §3.1 trigger-decl; §9 M22210/M22214
Issue: The grammar accepts procedure, event, and manual triggers, while M22214 says a workflow trigger MUST reference a procedure. The status of event and manual triggers is therefore contradictory.
Recommendation: Keep all three forms if they have distinct authorization and input contracts, and limit M22214 to rejecting a direct action trigger. Otherwise remove event and manual from the grammar and scope.
Q: How is a compensation step excluded from normal flow?
Status: Open
Source: RFC-0022 §3.1 step-list; §6.1–6.2
Issue: A compensation target must be a declared sequential step that runs only after failure, but every sequential step in the grammar belongs to the normal step list. No syntax or binding rule distinguishes a compensation-only step from one that also runs normally.
Recommendation: Add a dedicated compensation declaration or a named compensation block outside the normal step list, then specify type checking, authorization, and events for invocation.
Q: Can a later step depend on a variable produced by only one conditional branch?
Status: Open
Source: RFC-0022 §4.3, §5, §9 M22203, §10 invariant 1 Issue: The compiler emits only a warning when a later step depends on an output absent from the implicit or explicit else branch. Such a workflow can reach the later step without a value, contradicting the guarantee that all bindings resolve statically and no runtime data-flow type error occurs. Recommendation: Make M22203 an error whenever a required later input has no value on a reachable path, or require an explicit default binding on the missing branch.
Q: Can workflow-level onError: retry N recursively retry a failed step?
Status: Open
Source: RFC-0022 §3.1 error-policy; §6.1; §10 invariant 2
Issue: After a step-level retry N is exhausted, the RFC applies the workflow-level policy. That policy can itself be retry N, without a specified exhaustion continuation or a bound on repeated application. This conflicts with the claimed finite execution guarantee.
Recommendation: Define one bounded retry budget per policy and a terminal fail after exhaustion; disallow recursive reapplication of workflow-level retry to the same failure.
Q: Does finite execution include time spent at a human gate?
A: The bound applies to dispatched steps, not wall-clock duration. RFC-0022 §10 now says a human gate can remain open until a decision or a platform timeout outside this RFC.
Source: RFC-0022 §1 timeout exclusion; §4.5; §10 invariant 2 Issue: Static loop bounds make the number of step invocations finite, but a gate may wait indefinitely when no actor responds. The RFC excludes long-running timeout policy, yet says every workflow execution is provably finite. Recommendation: State that only the number of dispatched steps is bounded; wall-clock termination depends on gate decisions or a separately specified timeout policy.