Skip to content

Latest commit

 

History

History
151 lines (129 loc) · 9.69 KB

Integration.md

File metadata and controls

151 lines (129 loc) · 9.69 KB

CLI integration and distribution

The 1.4 integration branch starts at advanced filters 8d00757, draft PR #26. Its prerequisite is renderer 3c1e8a1a (PR #21). Transfer/RTF/PDF and parallel 1.2 graphs remain unchanged. The native executable has six root products and no external root dependencies. No public query library is exposed.

stream check|cat|array borrows the same job context as native parsing/writing. check waits for every required EOF before emitting {completed,inputRecords}. cat uses strict document/values/NDJSON framing and compact JSON+LF. array yields complete elements provisionally; closing syntax, trailing whitespace and EOF remain mandatory. A late error preserves an existing named output through exclusive staging, while previously accepted stdout bytes cannot be recalled.

schema check SCHEMA --resource URI=PATH compiles an explicit offline Draft2020-12 object or Boolean schema and emits completed metadata, without instance evaluation. Resource files, native meta-schemas, compilation and output share cumulative work. No reference fetch occurs. Known provider adapters return exit2/capabilityUnavailable: qualified helper declarations are absent at both renderer base and the observed transfer branch. Their mandatory authored success/rejection vectors remain blocked; a rejection test for unavailability is not a provider qualification success.

JSONHTML provides static (expanded, script-free) and explicitly interactive HTML. Text/Markdown use actual tab indentation and preserved exact number lexemes. No Apple PDF preparation or transfer-provider runtime is needed for these commands. The command-specific help and JSON version metadata enumerate only compiled capabilities; source-checkout revision remains unknown. Malformed UTF-8 in a filter file is compile exit3; malformed raw binding text remains input exit5; missing files remain IO exit7. Operational failures bypass language catches.

The real POSIX sink delegates progress to a private tested handoff loop. Short writes, readiness retries and foreign calls have checked progress/cancellation; zero progress is terminal IO. One handoff is outstanding, with no queued whole output. Caller/foreign instruction time and kernel latency are outside quotas. Finite work is monotonic; live ownership refunds never refund consumed work.

Distribution qualification copies the real release binary and schema bundle into a directory with spaces, executes outside the checkout, tests leading-dash input names and rejects a tampered sibling meta-schema. SwiftPM may retain a build-directory fallback; absence is tested separately only from clean reproduction with that original bundle unavailable. No assumption that copying only the binary is sufficient is made. Package-wide and selected-library claims stay distinct.

Tools/CLIQualification also has a qualification-only five-library mobile target. It imports JSON/Schema/Streaming/Patch/HTML, with no CLI, private filter, stdio or Apple export requirement. iOS/iPadOS share the iOS27 deployment declaration. WASI CLI feasibility and private wasm32 arithmetic outcomes are separate; a compiled kernel is not an argv/filesystem/signal/resource-qualified CLI.

Final-source receipts, exact acceptance statuses, archive/reproduction, installed consumers, platform observations and performance limitations accompany the return bundle. No merge, release tag, publication, PATH change or remote CI activation is performed by these adapters.

Native and platform milestone

Native, safety and core/HTML WASI lanes passed unchanged source manifest 68463696a494417a6509e7330cec0343a9199cb10f806cadfe4ff4d7506c9964. Native executed25 commands, debug/release297 functions/199 parameter cases. ASan matches those counts; TSan covers8 shared functions and strict-memory compilation has zero warnings. Baseline suites/cap/exclusion counts are retained. The affected CLI review lane then passed77 functions,53 foundation and36 integration subprocess cases,33 real site/report checks for16 caps, and105 C1-C4 vectors. The6 authored C5 rows have4 passes and2 blocked provider rows, not policy successes. An independent500-case bounded malformed-query replay had no crashes/timeouts.

The source review reproduced invalid presentation/framing being masked by input or compiler errors. Early argument checks and two real subprocess regressions now preserve usage exit2. Clean-source Git identity is recorded only when the queried top-level checkout is the actual source directory; ancestor Git is ignored. The predecessor clean928-file archive passed ten CLI-lane commands; its incidental ancestor Git SHA is explicitly non-authoritative. Final successor extraction uses archive/file hashes and isolated Git discovery.

WASI now has narrow WASILibc/entry guards, with native signal behavior unchanged. The CLI module compiles with matched Swift6.4.0 SDK. WasmKit0.3.1 executes exact format, Cartesian query, completed root-array stream, offline schema and static HTML commands. The first Node26.5.0 version execution with file-backed output failed IO exit7: actual poll_oneoff returned WASI errno28; tracing is qualification-only. The corrected module passes version/format/query/root-array/static HTML with pipe stdio. Explicit schema-file reading and regular-file stdout still fail poll with errno28 and exit7. Descriptor layout is recorded in the return matrix; these partial successes do not qualify all file/stdin/stdout adapters. No blocking fallback is used. CLI POSIX SIGINT bridging is absent on WASI, so these five command outcomes are a limited disposition, not full CLI qualification.

The preserved WASI lane executes220 core/HTML functions in debug/release and173 parameter cases,13 public consumer functions on each host,49 actual cap outcomes, 20 explicitly rooted downstream vectors, unchanged independent suites and native strict-memory compilation. The earlier104-count JSON harness was corrected upward to107 after actual discovery/passes; its failed receipt is retained. New CLI tests are compiled but are outside that core/HTML execution claim. Unsupported ASan/TSan WASI instrumentation retains its explicit original policy rather than a safety pass.

The selected five-library target compiled on iPhoneOS/iPhoneSimulator SDK27.2 with no CLI/filter target contamination. Release-native slow-sink medians were0.3144s for2000 and3.5305s for20000 records on a heavily loaded host; logical retention stayed about66.5KB and process RSS about7.6MB. These distinct observations support bounded logical pipeline state, not quiet-host throughput or hard RSS promises. Final-head hosted checks/reproduction and the return manifests bind the successor.

Checked resource successor

Clean installation of predecessor 9a773f3 reproduced a SIGTRAP when the sibling schema bundle and SwiftPM fallback were both absent. Its archive cf72bb6c… and failed receipt remain immutable and are not release qualification evidence. The successor replaces the generated trapping Bundle.module accessor with finite, checked app/framework/command lookup. Apple .bundle and Linux .resources layouts retain all nine size/SHA pins; a corrupt present bundle is not bypassed. Missing resources throw a schema metaSchema error. The actual installed consumer now requires exit5/no output for missing and tampered resources, with a positive sibling-resource case. One new Swift Testing function exercises missing lookup, shared work/retention, quota failure and in-loop cancellation.

Scripts/ci.py --lane native passed25 commands with298 root functions/199 parameter cases; safety passed ASan298/199, TSan8 and strict-memory/warnings-as-errors. The CLI adapter boundary is now53 foundation +37 integration subprocesses,77 CLI functions/26 parameter cases,33 checks across16 CLI metrics, and105 C1-C4 vectors. The resource-fix runtime/source manifest is c81e1eb3526a2f67546af9a1d906638ef1b506721390876d4bab8da5343619f8. Native, resumed safety and resumed WASI receipts record unchanged source inputs. WASI executes221 core/HTML functions in debug/release with173 parameter cases, all49 cap outcomes, public/rooted consumers and the unchanged pinned independent suite counts. Missing-resource and strict-memory probes pass. CLI WASI runtime support remains the separate limited disposition above. Both selected mobile SDK consumers passed on27.2 without CLI contamination.

Disk-full safety/WASI attempts and a SwiftPM lock-banner failure remain failed or incomplete records. Only completed caches owned by this CLI worktree were cleared; source and materialized receipts were preserved, and separate complete reruns passed. Resource-fix slow-sink release medians are0.0803s/2.7205s for2000/20000 records; peak logical retention is about66.5KB and measured RSS about7.5MB on a loaded host. There is no quiet-host throughput or hard RSS claim. Final-head hosted checks, source archive, clean extraction and acceptance closure are bound by the returned freeze/evidence records. Mandatory provider CL078/CL079 remain blocked.

Portable provider successor

The aded3c9 reconciliation is retained with35 passed hosted checks and its immutable source/evidence return. A distinct provider successor implements CL078/CL079 through public JSONSchema helpers; the historical unavailable results above are superseded by the provider contract, frozen profile and its new source-bound receipts. Native C5 vectors now have6 passes/0 blocked; the full authored corpus has111 successes. Unsupported/unverified declaration reports exit1; compilation, IO, resource and cancellation failures remain operational. Apple model/schema bridges remain a separate slice. No prior WASI/performance result is silently reused as qualification of new helper behavior. Current broad native, safety, docs, Linux and reproduction receipts accompany the successor return.