mirror of
https://github.com/hanzo-cpp/sdk.git
synced 2026-11-20 19:55:32 +00:00
This client read hanzoai/openapi's hanzo.yaml at branch `main` and carried no lock, so it named its document by something that moves and could not say which bytes it was a projection of. It now reads what hanzoai/cloud emits, at the commit .spec-lock names, digest-checked on every fetch — the same document and the same four-line receipt every other Hanzo client carries. This repo's own hanzo.yml had already declared that move; the script had not made it, so the lane and the script named two different documents. Nothing is lost, measured route by route: cloud's document is a strict SUPERSET of the projection — 2479 operations against 2456, with 0 routes only in hanzo.yaml and 23 only in cloud. The projection's one addition for this language, a `default` response injected into every operation that declares none, buys nothing here: a `default` with no content produces the same void return that no `responses` does, so both documents yield the identical 834 void-returning methods. Generated, that is 2502 methods against 2455 committed. The 7.24.0 generator floor is gone, and it was not wrong when written — both defects it named lived in schemas the document no longer carries (vector_NamedVectors under additionalProperties, framework_Document.docstatus). Re-measured at the locked commit, 7.14.0 and 7.24.0 emit the same library apart from the version string in each file's header and one doc comment on ConsoleSettings::getOidcConfig. So this repo is on the fleet's 7.14.0 pin and is no longer an exception on version. It stays out of sdks.yaml for one reason only, and that one is measured too: without templates/cpp-restsdk the library fails with 8 errors in ModelBase.h, and a template is a file that has to sit beside the invocation. One flag joins the invocation, and it is the fleet's own. o11y.GettableAgentCheckIn publishes integration_config beside integrationConfig and removed_at beside removedAt so older agents keep working; C++ camel-cases a property into its accessor, so each pair declared one method twice and the library did not compile. Renaming the accessor keeps both wire keys — all four are present in the emitted model, each under its own name — which is what the python, go, java, kotlin, ruby and php rows already do for the same two schemas. Four flows changed route, none of them because of this move: their old operations are untyped or absent in hanzo.yaml too, measured by regenerating from it, so the committed tree simply predated the change. hello, chat, money and tools move to the typed route for the same job; agent keeps its route and reads the run from the typed list, because the run POST declares no body and no response. Each says so in its own header, and the rule they follow is that an example has to demonstrate a call the document types. packageVersion leaves the invocation as inert: measured with a sentinel it reaches the generator's own README and no file under include/ or src/, so the library version in CMakeLists.txt is this repo's to choose and is coupled to nothing. Measured on aarch64 Ubuntu 24.04, g++ 13.3, generator 7.14.0, document 49389cf9: 2663 sources, library 0 errors, six examples 0 errors.