Repository navigation
Bracket file data in group named with filename. - #16
Merged
Merged
Conversation
bergie
added a commit
that referenced
this pull request
Apr 3, 2012
Bracket file data in group named with filename.
bergie
added a commit
that referenced
this pull request
Oct 9, 2026
Phase 2 of work document #28: RegistryProtocol, the Two-Step Cache server side, driven end-to-end through the runtime core in tests. The runtime consumes an application-supplied catalog — the same philosophy as the engine's application-supplied component registry (work documents #6/#16): signatures are ready at handoff, and the handler derives the wire state from them. A signature is exactly the 0x24 detail shape (declared kind plus the full port field set), valid with no implementation behind it, so manifest/sidecar-sourced stubs flow through unmodified. - 0x20 syncs answer 0x21 on hash match, 0x22 manifest otherwise — the runtime always answers (work document #4 update #18) - 0x23 details answer with only the requested known components; unknown names are omitted, and an empty response is an answer - 0x25 source writes and 0x27 installs map onto optional catalog hooks (source and signature are orthogonal, so a write never relabels a component); missing hooks surface as unsupported events - every catalog change re-derives sig_hash values and the registry_hash through the protocol package's canonical serializations Supporting changes: the runtime core now accepts async handlers (WebCrypto hashing is async), capturing rejections as error events like synchronous throws; @noflo/fbp-protocol re-exports its ComponentDetail/PortInfo/ ManifestEntry typedefs from the entrypoint so dependents can type against them.
bergie
added a commit
that referenced
this pull request
Oct 9, 2026
Phase 4 of work document #28: the 0x30/0x32 streaming telemetry surface, fed by a real NoFlo network. NetworkHost owns the network lifecycle: builds the NoFlo network from the runtime's graph with an injected component loader (platform-neutral, per work documents #6/#16), re-emits the network's EventTarget events so protocol handlers observe a network before, after, and across network recreations, and supports rebuild-on-discard. The engine gap it surfaces is documented: the 2.x engine consumes the graph at connect() and does not follow live graph changes, so CRDT operations applied while a network runs update the wire state but not the running network until the engine learns to follow graph deltas. TelemetryProtocol maps host events onto the 0x32 vocabulary: - data packets become DATA events with raw payloads; brackets become BEGIN_GROUP/END_GROUP with the group name - network start/stop become LIFECYCLE START/STOP codes - process errors become ERROR events with the exception string — except errors in the stub error class (default StubNotImplementedError), which become STUB_ERROR: the event type itself is the classification (#4 update #9) - payloads the wire cannot carry are frugalized to type-named placeholders so serialization can never take down the flush loop Flush policy is the runtime's physical decision: per-subscription buffers flush on timers at the effective cadence (the client's requested interval, bounded below by a configurable floor), with explicit flushing available for transports that prefer it. The protocol has no unsubscribe command; subscriptions live with their link, and dropContext tears them down on link close. The integration test streams a real network run — a real ComponentLoader, asComponent factory, graph with IIP — through the full pipeline to codec round-trips. @noflo/noflo re-exports the Network class, restoring 1.x surface parity that the host needed to type against.
bergie
added a commit
that referenced
this pull request
Oct 9, 2026
Phase 2 of work document #28: RegistryProtocol, the Two-Step Cache server side, driven end-to-end through the runtime core i 8000 n tests. The runtime consumes an application-supplied catalog — the same philosophy as the engine's application-supplied component registry (work documents #6/#16): signatures are ready at handoff, and the handler derives the wire state from them. A signature is exactly the 0x24 detail shape (declared kind plus the full port field set), valid with no implementation behind it, so manifest/sidecar-sourced stubs flow through unmodified. - 0x20 syncs answer 0x21 on hash match, 0x22 manifest otherwise — the runtime always answers (work document #4 update #18) - 0x23 details answer with only the requested known components; unknown names are omitted, and an empty response is an answer - 0x25 source writes and 0x27 installs map onto optional catalog hooks (source and signature are orthogonal, so a write never relabels a component); missing hooks surface as unsupported events - every catalog change re-derives sig_hash values and the registry_hash through the protocol package's canonical serializations Supporting changes: the runtime core now accepts async handlers (WebCrypto hashing is async), capturing rejections as error events like synchronous throws; @noflo/fbp-protocol re-exports its ComponentDetail/PortInfo/ ManifestEntry typedefs from the entrypoint so dependents can type against them.
bergie
added a commit
that referenced
this pull request
Oct 9, 2026
Phase 4 of work document #28: the 0x30/0x32 streaming telemetry surface, fed by a real NoFlo network. NetworkHost owns the network lifecycle: builds the NoFlo network from the runtime's graph with an injected component loader (platform-neutral, per work documents #6/#16), re-emits the network's EventTarget events so protocol handlers observe a network before, after, and across network recreations, and supports rebuild-on-discard. The engine gap it surfaces is documented: the 2.x engine consumes the graph at connect() and does not follow live graph changes, so CRDT operations applied while a network runs update the wire state but not the running network until the engine learns to follow graph deltas. TelemetryProtocol maps host events onto the 0x32 vocabulary: - data packets become DATA events with raw payloads; brackets become BEGIN_GROUP/END_GROUP with the group name - network start/stop become LIFECYCLE START/STOP codes - process errors become ERROR events with the exception string — except errors in the stub error class (default StubNotImplementedError), which become STUB_ERROR: the event type itself is the classification (#4 update #9) - payloads the wire cannot carry are frugalized to type-named placeholders so serialization can never take down the flush loop Flush policy is the runtime's physical decision: per-subscription buffers flush on timers at the effective cadence (the client's requested interval, bounded below by a configurable floor), with explicit flushing available for transports that prefer it. The protocol has no unsubscribe command; subscriptions live with their link, and dropContext tears them down on link close. The integration test streams a real network run — a real ComponentLoader, asComponent factory, graph with IIP — through the full pipeline to codec round-trips. @noflo/noflo re-exports the Network class, restoring 1.x surface parity that the host needed to type against.
bergie
added a commit
that referenced
this pull request
Oct 10, 2026
Networks no longer fall back to graphMetadata().componentLoader when no loader is passed via options; the deprecated path warned since the #16 registry work. Loaders come from network options or the application registry.
bergie
added a commit
that referenced
this pull request
Oct 10, 2026
Since the registry-only loader (#16) the catalog is populated synchronously at construction and listComponents() is a plain Promise.resolve -- awaiting it before connect() is a vestige of the fbp-manifest discovery era. - NoFlo.createNetwork and Subgraph.createNetwork no longer call it; the subgraph keeps a microtask hop so the 'network' lifecycle event stays observable to listeners attached right after construction - Specs: drop the warm-up calls from before() hooks (the ones asserting the catalog API itself keep them)
AsherBond
pushed a commit
to AsherBond/noflo
that referenced
this pull request
Oct 12, 2026
AsherBond
pushed a commit
to AsherBond/noflo
that referenced
this pull request
Oct 12, 2026
Work document noflo#16, step 1 of the registry-only core split. The function-wrapping helper was the only user of the get-function-params dependency, which now leaves noflo core entirely. - New workspace package @noflo/as-component (EUPL-1.2) holding asComponent and its spec; takes noflo as a peer dependency and is never imported back by core - noflo drops the noflo.asComponent export and the get-function-params dependency
AsherBond
pushed a commit
to AsherBond/noflo
that referenced
this pull request
Oct 12, 2026
Work document noflo#16, step 2 of the registry-only core split. The Node.js discovery machinery (fbp-manifest listing, fbp.json cache, on-the-fly TypeScript transpilation, CommonJS evaluation, custom noflo.loader plugins, source storage) becomes a standalone ComponentRegistry implementation. - createNodeModulesRegistry(baseDir, options) discovers components and resolves to a ready registry: list() is synchronous, get(name) stays async per the registry contract - Discovered graph files register as pre-parsed GraphModel instances - setSource stores, transpiles, and evaluates source, dispatching a change event; getSource includes fbp-spec tests where discovered - noflo-cache-preheat bin moves here; loadGraphFile/saveGraphFile move here from core (EUPL-1.2, greenfield package)
AsherBond
pushed a commit
to AsherBond/noflo
that referenced
this pull request
Oct 12, 2026
The component class wrapping child networks was Graph in a file named Graph.js, which collides conceptually with the @noflo/graph GraphModel. Renames the class and file to Subgraph; the loader catalog entry that classic discovery registers follows suit. Work document noflo#16 prepares this by making the loader instantiate the wrapper directly; this rename is the mechanical part, landing first so the registry-only core change stays a pure semantic step.
AsherBond
pushed a commit
to AsherBond/noflo
that referenced
this pull request
Oct 12, 2026
Work document noflo#16, final step. Deletes the whole NodeJs-loader concept from core: no path strings, no source evaluation, no CommonJS, no transpilation, no discovery. Core consumes an application-supplied ComponentRegistry plus manual registerComponent calls: - ComponentLoader constructor is options/registry-only; the registry catalog is read synchronously at construction (registries are ready at handoff), listComponents stays as the noflo#8 promise accessor - registry EventTarget change/invalidate events keep the loader cache current for live systems (FBP runtimes, IDE sessions) - setSource/getSource/getLanguages/dynamicLoad and the sourcesForComponents/specsForComponents scratch fields leave the loader; source storage is registry API only - baseDir exits the loader, networks, and asCallback: pass componentLoader or registry instead; subgraph child networks inherit the parent loader - loadGraph instantiates the Subgraph wrapper directly; the Graph catalog entry is gone and the runtime graph inport on directly-instantiated subgraphs warns deprecated - isGraph recognizes live GraphModels, FBP JSON, and legacy processes/connections JSON shapes - fbp-manifest, @noflo/fbp, and the noflo-cache-preheat bin leave the package; core dependencies are @noflo/graph and debug - spec fixtures convert from .fbp DSL to committed FBP JSON so the parser leaves the devDependency list too - fbp-spec-runner compiles suites through @noflo/loader-node discovery of the project under test
bergie
added a commit
that referenced
this pull request
Oct 19, 2026
A component with no error port declared that calls done(err) escalates through a load-bearing chain: error() throws, fireProcess's catch answers with its own sendDone(e) which throws again from inside the catch, the synchronous listener-throw propagation rejects the delivering edge's write, and the socket's error side-channel turns it into the network's process-error — or the loud async throw when no socket error listener exists (standalone). Verified empirically and held by a network-level pin: the chain is exactly what the no-silent-loss invariant needs for components without an error leg, and a refactor of the catch could sever it silently. Work document #48 update #16.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
If you're sending a bunch of filenames to ReadFile, it seems useful to be able to keep track of which files the data came from. If I'm overlooking a better way of handling this, please let me know.