Do not initalize port groups in Component by rybesh · Pull Request #19 · noflo/noflo

rybesh · 2012-04-01T06:46:23Z

Initializing inPorts and outPorts in the base Component class is dangerous, because all components will then share the same inPorts and outPorts groups, unless they explicitly create new objects to initialize these in their constructors. If they simply assign to the existing objects (as was being done in ReadFile), those assignments will affect all components, and components may end up with spurious ports being created for them. Better to force components to initialize these themselves.

Note that this bug affected the two components I submitted pull requests for earlier (ReadDir and Stat), so this pull request also includes those two.

Only return paths that actually exist.

The library naming conventions are consumed by hosts classifying component names against their own project namespace (work document #19 review: the stub upload's reboot-stable identity is the project's own library name). Doc and export only; the function is unchanged.

…oflo#19) The canonical class the registry contract hands back for top-down design (noflo#593): declared ports render and validate graphs, outports are declared but never sourced, bracket forwarding is off, control inports are inert gate ports, and brackets are absorbed — consumed, never emitted. Data packets on triggering inports raise the dedicated StubNotImplementedError (carrying componentName, wire-classifiable as 0x09 STUB_ERROR) through output.error; the opt-in onPacket: "drop" policy consumes and discards instead. Policy throw with zero triggering inports is rejected at construction. StubComponent.fromSignature maps both signature entry paths onto plain port options: the full manifest field set and the narrower on-disk sidecar shape noflo-ui materializes. The Component base gains isStub() (false; the stub class overrides true), so consumers label "not yet implemented" from authoritative runtime information. The hand-rolled noflo#593 dummy in the ComponentLoader spec is replaced with the canonical class.

watch/replay printed post-0x07 flowtrace events as raw numbers; a stub component's 0x09 STUB_ERROR rendered as 9 instead of its name. The table now covers every EVENT_TYPE entry (work document noflo#19's stub surfacing plus the breakpoint, capacity, connection, activation, and stranded events already in the shared vocabulary).

components/<name>.json — the per-component signature JSON noflo-ui materializes into a project workdir — is now a distinct file kind in the components directory scan (work documents noflo#19 and noflo#24): classified by shape (inports and outports both arrays), registered as a StubComponent factory via StubComponent.fromSignature, one fresh instance per load. A same-named implementation source wins; the sidecar is dropped as documentation of the declared signature, and the entry graduates when the implementation arrives later. Stray JSON keeps the pre-sidecar ignore behavior. Fixtures: a stub-only entry, an implemented sibling with its sidecar, and a non-sidecar JSON.

…cars The host's component catalog gains the 0x25/0x26 source hooks (work document noflo#19). A source payload classifying as a signature declaration — the noflo-ui sidecar shape, inports/outports arrays; component source and signature JSON grammars do not overlap — materializes as a canonical components/<name>.json sidecar in the project workdir, registers as a StubComponent, and signals the catalog change: a client uploads a stub, the sidecar persists it, discovery re-derives it on every boot. Signature uploads over existing implementations are rejected (implementation wins); other payloads store as source through the registry's setSource, implementing a stub in place. 0x26 reads round-trip a stub's sidecar. componentDetail pins the stub kind through the core isStub().

The COMP_WRITE payload may also be a signature declaration (work document noflo#19): the on-disk sidecar shape noflo-ui materializes, uploading a stub. Classification is the runtime catalog's, not the wire's — the source string stays opaque, and component-source and signature-JSON grammars do not overlap. Documentation only: no codec or frame-shape change to the frozen wire format.

stubs filters the synced Two-Step Cache catalog to entries whose declared kind is stub — the runtime's unimplemented components, full signatures included. stub set uploads a signature sidecar as a stub: the payload rides the ordinary 0x25 source write as a JSON signature declaration, validated client-side so a non-signature file fails locally. Both surface as CLI commands with --json output.

…ntract Doc-only: the RuntimeCatalog.writeSource hook description mentions the 0x25 signature-declaration convention (work document noflo#19) — the payload may be the on-disk sidecar shape noflo-ui materializes, classified by the catalog. No code change: the 0x25 handler already delegates and refreshes.

rybesh added 3 commits April 1, 2012 02:36

Added ReadDir component.

43cceab

Added file status component.

9b0b9da

Do not initalize port groups in Component.

2a7922d

bergie merged commit 2a7922d into noflo:master Apr 3, 2012

AsherBond pushed a commit to AsherBond/noflo that referenced this pull request Oct 12, 2026

Merge pull request noflo#19 from ensonic/master

63973e2

Only return paths that actually exist.

Navigation Menu

Repository navigation

Do not initalize port groups in Component - #19

Do not initalize port groups in Component#19
bergie merged 3 commits into
noflo:masterfrom
rybesh:componentfix

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants