Skip to content

chore: remove obsolete DAO tooling and simplify dev workflow (#380 part 1) - #388

Merged
oglassdev merged 6 commits into
mainfrom
cleanup/archive-tooling
Sep 24, 2026
Merged

oglassdev merged 6 commits into
mainfrom
cleanup/archive-tooling

Conversation

@oglassdev

Copy link
Copy Markdown
Owner

Part 1 of #380: remove obsolete non-parser tooling and fix small tooling bugs. No library or CLI behavior changes; crates/jet3 is untouched.

Net: 389 files, +974 / -87,242 lines against main (9638d38). Everything deleted is preserved at the annotated tag oracle-archive-2026-09, so the old DAO runs stay reproducible there.

Removed

  • DAO harness generations G1-G3 (~72.7k lines under oracle/, scripts/, .github/):
    • G1, hosted protocol v1.2 / A9: Invoke-Dao*V12.ps1, Invoke-DaoAllocationA9.ps1, dao_*_diff.py, dao_allocation_a9.py, hosted_write_reanalysis.py, remote/Remote.Process.ps1, protocol/v1_2/WRITE.md, their acquisition plans and tests, plus the 9 windows-dao-*.yml workflows.
    • G2, local dev jobs: scripts/windows-dao-dev.py, oracle/windows-dao/scripts/dev/*, their analyzers (bootstrap_*, schema_generalization, lvprop_null*, multiple_indexes, definition_continuation, extended_names, multi_table_create), test_windows_dao_dev_contract.py and the 22 windows-dev-* recipes.
    • G3, pinned-plan phased suites that each build their own SSH command (field_update.py pattern): about 45 .py and 37 .ps1 files, with their plans and tests. Also the unreferenced relationship_graph_*_auto.ps1 / _progress.ps1 variants.
  • testkit (~2.7k lines): classifier_snapshot and its binary, docs/validation/stage1-classifier-snapshot.json, the six hosted-workflow *_fixture.rs producers, all 8 binaries, and the unused tempfile dev-dependency.
  • tools / fixtures / validation: the preservation-diff verifier, the private ETS3000 external-corpus tooling and docs, the empty fixtures/ tree and its schema.
  • Fuzz / bench attestation (~7.6k lines): fuzz_campaign/build_identity/evidence/process, the registry, seed manifest and schema, and the bench scripts, manifest, schemas and baselines. Targets, seed corpora and the three .rs benches are unchanged.
  • Docs (~1k lines): docs/architecture/, EVIDENCE.md and ACCEPTANCE.md (merged into docs/validation/README.md), and the ~540-line V1_SCOPE checkpoint log (PROVENANCE is the record).

Kept, and why

  • scripts/dao-check.py and all 12 registry suites, plus every prepare/compare suite run via scripts/windows-dao-ps.py (all suites added from 2026-09-14 onward, EXP-0222+), system_catalog.py and multi_level_index_structure.py.
  • Helpers that kept suites imported from deleted files moved into small modules: dao_common.py (digest/identity/canonical/normalized, validate_environment, leaf_entries, index_tree, load_catalog) and catalog_keys.py (catalog name keys and LvProp decoding extracted from schema_generalization.py, with its tests moved into test_catalog_keys.py).
  • Differs from the plan: the protocol/v1_2/ contract (inventories, schemas, TSV vectors, README) and validate_protocol_v1_2.py / protocol_validation.py / build_v1_2_inventory.py stay. jet3-cli snapshot embeds those inventories (--inventory row-update|...), testkit's value tests use the TSV vectors, and the CLI snapshot tests validate output with that validator. Deleting them would change CLI behavior. The six inventory constants moved into jet3-testkit/src/scenarios.rs, and write_fixture.rs stays because a CLI snapshot test uses it.
  • field_update.ps1 stays unchanged. The kept producers load its helper functions and pin its hash.
  • oracle/windows-dao/README.md is rewritten: how to run the kept suites, plus an EXP-range to archived-harness table.

Tooling fixes

  • just ready = fmt-check lint test doc source-size contracts, so each check runs once. scripts/acceptance.sh is deleted, which removes its hard-coded 1.96.0 toolchain and the repeated quick run; rust-toolchain.toml covers the toolchain. contracts runs the support-matrix validation, the protocol inventory and schema checks, and the tools and oracle Python tests.
  • New recipes: just fuzz [seconds] (loops cargo fuzz run into fuzz/target/corpus/<t> so checked seeds stay untouched) and just bench.
  • CI: clippy added to the formatting job; Miri job dropped (the crate forbids unsafe), along with Miri in the mise hook; the fuzz job is a plain 30 s-per-target loop; the bench job runs cargo bench -- --test.
  • Support matrix: evidence paths that pointed at deleted files are dropped; docs/PROVENANCE.md remains cited.
  • CLI: the top-level --help usage now lists all seven commands (it only listed probe). The duplicate 256 MiB input-limit constant is deduplicated, and the mutate/schema/inspect tests share tests/common/mod.rs. The two budget() helpers were not merged: inspect uses a 256 MiB read limit and values uses the library defaults, so merging them would change behavior.

Checks

  • just ready passes (fmt, clippy -D warnings, 823 + CLI/testkit tests, rustdoc -D warnings, source size, contracts, 56 oracle and 4 tools Python tests).
  • py_compile plus an import of every kept oracle module passes.
  • cargo bench --manifest-path benches/Cargo.toml -- --test passes.
  • just fuzz was not run locally because cargo-fuzz isn't installed here. The CI fuzz job runs the same loop.

🤖 Generated with Claude Code

Drop the hosted protocol v1.2 / A9 harnesses and their nine workflows, the
windows-dao-dev.py discovery jobs, and the pinned-plan phased suites that
each built their own SSH command. Kept suites (dao-check.py registry and the
prepare/compare suites run through windows-dao-ps.py) now import their shared
helpers from dao_common.py and catalog_keys.py. The protocol v1.2 contract
stays because jet3-cli snapshot embeds its inventories.

Everything removed is reproducible at tag oracle-archive-2026-09.
Drop the stage-1 classifier snapshot, the hosted-workflow fixture producers
and their binaries, the preservation-diff verifier, the private external
corpus tooling and the empty fixtures tree with its schema. The six scenario
inventories used by jet3-cli snapshot move to testkit's scenarios module;
write_fixture stays because the CLI snapshot test uses it.
Replace the fuzz campaign/attestation tooling with plain cargo fuzz runs
(just fuzz / a CI loop) over the existing targets and seed corpora, and the
benchmark manifest/baseline scaffolding with plain cargo bench. just ready
now runs each check once instead of repeating them through acceptance.sh,
which is removed. CI gains clippy and drops the Miri job (the crate forbids
unsafe).
The top-level --help usage now names all seven commands before the
per-command sections. inspect reuses the crate-wide 256 MiB input limit and
the mutate, schema and inspect tests share one request helper.
Merge the validation README, evidence vocabulary and acceptance gates into
one short docs/validation/README.md, drop docs/architecture (its two
cross-cutting rules move to the root README), and cut V1_SCOPE down to
deliverables, exclusions and remaining work; PROVENANCE remains the record
of the old checkpoint log.
@oglassdev
oglassdev merged commit f084e8c into main Sep 24, 2026
10 checks passed
@oglassdev
oglassdev deleted the cleanup/archive-tooling branch September 24, 2026 04:45
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant