Lightning.Adaptors.NPM (Lightning v2.19.0)

Production implementation of Lightning.Adaptors.Strategy that talks to the public NPM registry and the OpenFn adaptors monorepo on GitHub.

• Lightning.Adaptors.Strategy.list_adaptors/0 merges the @openfn org's package listing with the search API's version lookup. See Lightning.Adaptors.NPM.Registry for why this is two calls, not one. A listing with no @openfn/language-* names is an error, not an empty catalogue.
• Lightning.Adaptors.Strategy.fetch_adaptor/1 fetches the packument and the latest version's schema from jsDelivr. Icon fields are not stamped here. The Scheduler joins them on after a bulk Lightning.Adaptors.Strategy.fetch_icons/1 pass.
• Lightning.Adaptors.Strategy.fetch_icon/2 is a single raw GET against raw.githubusercontent.com, used when the Store misses an icon on disk.
• Lightning.Adaptors.Strategy.fetch_icons/1 fans out one GET per (name, shape) and sends :prior_etags as If-None-Match.

HTTP

The HTTP work lives in three sub-modules, each with its own Tesla client and base URL: Lightning.Adaptors.NPM.Registry, Lightning.Adaptors.NPM.Schema and Lightning.Adaptors.NPM.GitHub.

Every request is single-shot and bounded by http_timeout. There is no retry, backoff or circuit breaker. A transient failure (5xx, timeout, nxdomain) of the primary request (the packument for fetch_adaptor/1, the org listing for list_adaptors/0 and fetch_icons/1) surfaces as {:error, term()} unchanged, and a failed schema fetch inside fetch_adaptor/1 as {:error, {:schema_fetch_failed, reason}}. Icon fetches inside fetch_icons/1 are best-effort instead: a miss degrades to an absent icon shape rather than failing the batch.

Configuration

The sub-modules read :registry_url, :jsdelivr_url, :github_url, :github_ref and :http_timeout from Lightning.Adaptors.Config.strategy_opts(Lightning.Adaptors.NPM). All three share this module's config key rather than having their own, and each key has a default so the module works with no Application env set.