18 KiB

Raw Blame History

Repository Landing Plan

Audit Snapshot

This plan is based on the local repository state audited on 2026-08-18. No refs, worktrees, indexes, stashes, or remote state were changed.

The integration baseline is:

main        1abd47609b161c4b9fd460274dc9123755211436
gitea/main  1abd47609b161c4b9fd460274dc9123755211436

dev/main and srht/main both stop at a00e9c0e648dbba29e6dc4e4cc51267b24aa478c, 241 commits behind the integration baseline. They must not be used as PR bases.

The audit covered:

all seven registered worktrees;
168 local or cached remote-tracking refs with post-2025 unmerged tips;
patch-equivalent and rebased histories;
eight stash entries;
reflogs and 483 unreachable commits dated in 2026;
old merged topic branches and abandoned provider branches.

No recovery-worthy application patch exists only in a stash or unreachable object. Useful abandoned work remains reachable through a retained branch.

Current forge state was not queried. PR 92 and Tractor PRs 475/490 are mentioned by local plan artifacts, but their current head, base, and status must be refreshed before branch preparation.

Governing Rules

Create clean extraction branches from the then-current main; do not merge any aggregate development branch wholesale.
Preserve source commit order inside each lane, but consolidate fixup, formatting, and test-repair commits where noted.
Remove ai/prompt-io/**, generated provider links, local skill deployment, plan-only commits, editable dependency pins, and local lockfile paths from product PRs.
Keep source and backup refs until the extracted PR is accepted and its resulting tree is checked against the intended source tree.
Refresh forge metadata and dependency SHAs immediately before preparing each branch. Cached remote-tracking refs are not proof of current PR state.
Land independent small fixes while the service/datad chain is blocked.

Landing Sequence

Phase 0: Refresh External State

Before changing history:

Refresh main and identify current open/closed PRs for every retained Gitea branch.
Verify whether documented piker PR 92 still points at wkt/fix_broadcast_consumers / 4af885d9.
Verify Tractor PR 475 and PR 490, record their accepted OIDs, and determine the first released or stable Tractor revision containing the required service and address APIs.
Select the PyQtGraph source policy: official 0.14, upstream, or the Pikers fork. Do not let an incidental lockfile decide this.
Snapshot branch-to-PR mappings before deleting or rewriting any topic ref.

Phase 1: Independent Low-Risk Patches

These do not depend on the service/datad stack and should land first.

1A. Accounting Synopsis

Extract 45d2b53bf912d006f486c644f2add2e313ec6fcc as a one-commit documentation PR touching only piker/accounting/README.rst.

1B. Symcache Correctness

Create one two-commit PR in this order:

a3ed73a40d42db4bbc3032a296413d33b12eab99  strip TOML-incompatible None
5f392ff0bb0281422d4fdb8e4ebd26e70b04a54a  persist actor-local cache

Add focused serialization and repeated-open tests. The first patch comes from an old base; the second is a direct child of the audited main.

1C. Small IB Correctness Fixes

Use one small IB PR, preserving this order:

5df511f7d2de7712156a3556e1d67986706408c0  handle stringified errno
de82df727d2e828c5759e32aa758f7f1f7d498e1  expiry and venue helpers
79be47635f5e4eef81374e63f774e62a1cec5931  unknown venue calendars
170c95da2815bc0b1fd48e4332f7ff4b4ccaeb99  ledger exchange fallback

The first and fourth patches have compatible main preimages. Keep the venue pair together and add unknown-calendar and exchange-fallback tests.

1D. Qt Graceful SIGINT

Extract only 8dee35cfa7fbbf9eb614dd709ef86632649a15b4^!; never use its inherited 67-commit branch range. Require focused Qt/KBI shutdown coverage.

1E. DPI and Font Auto-Calculation

Replace the DPI branch swarm with one five-commit PR sourced from the clean range 1abd4760..768b6f31:

Consolidate cdb0e841, 574fb80d, 3b0c27ec, 5c7f00a3, and e600d61e into the font-size calculation and test commit.
Consolidate 5b1c80a8 and ccc5a745 into Qt screen diagnostics.
Keep 3f77180b as the physical-PPI helper.
Consolidate 88f66bae, 17e7232d, 2e42d7e1, and ce8ad59f into global font zoom and axis relayout.
Keep 768b6f31 as the macOS axis gate.

Validate headless Qt, repeated Qt6 zoom, and macOS axis behavior. Evaluate ee353099 separately for its environment-variable finding; do not silently fold it into the feature.

Phase 2: Graphics and Gap Foundation

The current gap_annotator branch mixes DPI, dependency, data hygiene, and annotation work. Replace it rather than merging its 24 commits.

2A. PyQtGraph Policy

After the Phase 0 decision, make one dependency-only PR and regenerate the lock on current main. Candidate source intents are:

34281459518d838b85f3eaa9a9eae7d6f30c2ca0  official pyqtgraph 0.14
edecdb3ede62104bf7f881330786dd545d07e799  upstream development revision
d98c1706                                         later official-pin variant

These are alternatives, not a stack.

2B. Gap Annotation Core

Consolidate 8c08ddd3 and 073176a4 into one batch-rendering/client-protocol commit, while incorporating the later ownership, indexing, cleanup, routing, and tests from Phase 6. Remove forced profiling and the editable Tractor pin f9979956.

Keep unrelated canonical-line changes outside this PR:

b87710e9  null-bar tolerance
e30957b6  cyclic Flume.feed removal
31859e9d  crypto-future symbol handling
392ea616  obsolete poetry.lock removal

Review each as a small independent data/dependency patch rather than making it a hidden prerequisite.

Phase 3: Runtime and Service Foundation

This is the first blocking chain. Land in the following order.

3A. Service Manager to Tractor

Use only the clean cached Gitea range:

1abd47609b161c4b9fd460274dc9123755211436
    ..8a02f6ea39c11ebc463a5b0a33dbfd5d6492a26e

Replace the hilevel_serman development pin with the verified Tractor OID. Settle self-service cancellation semantics before accepting test changes.

3B. Multi-Address Runtime

Extract only:

604e5fcf9c73642804115074d80ff688da39325a  use tractor.Address
5466acb7647b762db3688ec424e1fff618ca062a  bind-address and registry endpoints

Add tuple/wrapped-address, multi-bind, and registry-election tests. Do not merge multiaddrs or the 54-commit inherited branch range.

3C. Current Runtime/Test Repair

Land 4485f2b9^..bc6e18d7 on 3A and 3B. Remove prompt provenance and avoid the private tractor.discovery._addr.wrap_address() API if a public equivalent exists. Do not make broader accepted cancellation outcomes hide regressions.

Phase 4: Datad and Brokerd Split

Split the original aggregate datad_service line into reviewable PRs.

4A. Daemon Contracts and Datad Foundation

119d2c049556f240e6b457d08c5c6d564ece011e
6418121923f33f17ac7283c23ffe984af1f8f98f

4B. EMS and Feed Ownership Split

59d5d9a66d5b71d49cb87d514d0b2436190d330a
eee19de0901a8f7367885f6c8e8b8af18626776c
f15f8178a3fd9af223673666db194e6f9234ca7e

Replace live-Kraken-dependent startup tests rather than increasing timeouts. Verify capabilities, actor IDs, client-ID offsets, and teardown.

4C. Datad Logging and IB Contract Qualification

Keep datad logging with datad:

ba32f286b9651d6f65a4a17d9e9992a8a5b267c0

Use a separate IB follow-up in this order:

456c6a55677febd0e61bf394a1acc835c1941223
b0766764f03b921b1fe05bbe8586db698c8d2c03

The contract qualification patches repair caches that no longer warm in the same process after the datad split.

Omit f084e899 and b73300c8: they are design/provenance and AI-skill-only commits, not runtime dependencies.

Phase 5: Post-Datad Concurrency Fixes

5A. IB MethodProxy Correlation

Combine/rework these as one focused request-correlation series:

0404e4230ef00cbde21cfa2f73006b4dd1211fcd  request ID correlation
2aa66c2cf6c76b8e270af9367d27ca16f63e7024  response ownership/routing

Preserve the final concurrency, cancellation, and EOF tests. Strip skill and prompt files.

5B. Broadcast Consumer Ownership

Preserve source chronology but submit small adapter-scoped PRs where forge review permits:

12b0d66e854247e92cb5a8668be71d816c72d856  EMS dark quotes
b691422b2678e1e7b3d4b7e1d70c25d6c644c6d3  generic cached feeds
ddac3553a3bd4e2c29a7d2188dffb2c10ee065da  Deribit
4af885d90f6e4bcfa963f2f7a1d40436d3c928f3  IB

Each source commit has a dedicated test. If local plan data is current, these are represented by piker PR 92; verify before replacing that PR.

Phase 6: Backfill, Storage, and Chart Gaps

6A. Backfill and NativeDB Correctness

Rebuild the source sequence without plan/provenance commits:

976db6c3  normalize gaps after sorted dedupe
931f22d2  drop orphaned backfill events
c60a5a45  report non-positive timestamp steps
b000db94  atomic non-destructive NativeDB backfills
689df816  remove redundant writer locks
19c6d98e  normalize provider OHLCV
ed85721c  clean test-owned SHM leaks
1f309170  repair legacy timestamps
d7d537d5  raw OHLCV audit
922a5df8  qualification workflow
ce33deb6  IB UTC history syntax
d359b1cb  repair first-slot sentinels
ad6c560f  guard invalid SHM snapshots
0846cbd4  repair IB startup gaps
74ea2152  expose storage state
f0f3c806  simplify store discovery
06d5ea5d  repair sparse IB prepends

Split this into NativeDB/storage core and IB-history adapter PRs if the intermediate tree remains testable. Omit 99fb9d2e, 6913e80d, and 84a6d47b; they are plans/profiling notes. Strip ai/prompt-io/**, skill files, and plans/** from product commits.

Known blockers are deterministic NativeDB/SHM tests, live-provider qualification, exact-null repair, venue classification, and explicit history revision publication.

6B. Chart-Local Gap Overlays

Land this final visual ownership/test sequence:

f65df6a138cb8dc2119feb421c122a2c8ec2b89e
78982d8280870be6e2f7aa183e86f06265b63773
711666f9c831f9ae46b3a615c7f4c39beb1de140
b738c9f4f7a67da0da64b5acc5adcb3f1a10388b
335190a4e3ff5712626288ac5e81693b9646c140

Omit 09ddcf50, which is a pytest-qt plan only. Preserve the real actor test. History-revision-triggered overlay refresh remains follow-up work and should not block the ownership correction.

Phase 7: Independent Provider and UI Follow-Ups

These can proceed in parallel once their direct prerequisites settle.

7A. Kraken Account and Token Repair

Use two focused PRs:

ebc5bbd4 + 923b4de2                 account aliases/config loading
5387538b + 709269fc + Kraken hunks
from 8299c658                       stale token/session refresh

Add account-default, expiry, reconnect, and stale-token tests.

7B. EMS Edit-State Repair

Extract 8cefc1bd, e1cd3fd9, and the EMS portion of 9f7c38a3. Add rapid-edit/cancel tests. Reimplement the nested order-mode lifetime patch from e05c2589 only after this PR, with explicit cancellation and context-exit coverage.

7C. General Broker Error Schema

Preserve this order:

7f8a198c98528feb8c0981245a64288b84b946fd
84c742a885e8ca374b98fea5b84858bd2c3c38c4
bf43036fe10513f5532cacf24d4f381944f9aec8

Adapt IPC exception registration to the accepted Tractor API and remove unrelated IB hunks.

7D. Annotation Refresh Positioning

Rebase 555c2fc27e4dca0215454611b169232758955b8f and add zoom, range, and pan tests. This is unique but should follow the DPI/gap graphics lanes.

Phase 8: Build, Tooling, and Documentation

8A. Nix Dev-Shell Refresh

Apply in order and regenerate the final lock after dependency decisions:

f6dc218a8e6028489a07e1c1fb4971d9f2fbda58
3da8900f681a15fd4ea9f4728ff788633b54275f
ceb806c1365497242ab025681f41f9c3d37078ab
58ffe487759ca6816f7224488b99966f4acaf6a5

8B. Run-Tests Harness

Create one reproducible repository-tooling PR from the functional files in c83983b0 and e766c158. Keep the Piker py313, Nix, and pytest -p no:xonsh reference in-repo. Do not rely on ignored external symlinks in a fresh clone.

8C. Optional Dependencies and Lint Policy

Reapply the VisiData intent from b55e7e1d to current pyproject.toml and regenerate the lock; do not cherry-pick its stale lockfile.

Treat c86e69b2 and cd23021b as a Ruff-policy proposal. Run a full-tree violation audit before accepting it. Only then consider the dependent lint skill in bb2ea50f.

8D. NGI Proposal Documents

Use ccb6f191, 7060faa5, and b05292b0 as a documentation-only series. Omit ca100290 unless its generated summary is intentionally reviewed and accepted.

Deferred but Preserved Work

Deribit Backend

Preserve 5e371f1d..b209512e as the canonical 13-commit restoration source. It is high-value but 249 commits behind, lacks focused tests, contains whitespace errors, and may target a drifted API. Validate and reconstruct it; never merge the old branch directly.

Keep the max-pain range b209512e..b9321dbb behind the backend restoration. Defer OI/native-storage commits 8f1e082c and f756f0fd until both Deribit and NativeDB APIs settle.

Symbol Search

Do not land fixing_piker_search / 376a900b. It pauses almost every backend and mixes registry initialization. Salvage only the brokermod.name and nursery-argument correction into a new tested patch after MethodProxy work.

Other Deferred Work

Recreate private testing helpers from 4f2fd73b; do not cherry-pick its 73-commit aggregate ancestry.
Evaluate legacy Binance HTTPX commits 01a43763 and 0c61b0d9 against the current API before deciding whether to drop them.
Preserve broad macOS branches until their unique geometry, SHM, config, script, dependency, and documentation changes receive a separate audit.
Drop decimal EMS TOSQUASH remnants unless a current regression proves they are still required.

Worktree Disposition

All worktrees are read-only until their source lanes land.

Worktree branch	Head	State	Disposition
`chart_local_gapper`	`09ddcf50`	dirty root worktree, one plan commit above chart source	retain; never reset; omit plan commit
`wkt/backfiller_deep_fixes`	`f0f3c806`	clean	retain as backfill source
`wkt/fix_broadcast_consumers`	`4af885d9`	untracked PR-message files	retain through PR 92 verification
`wkt/flake_update`	`ceb806c1`	untracked skill config	retain as Nix source
`wkt/run_tests_skill`	`c83983b0`	untracked skill config	retain as harness source
`wkt/chart_local_gap_overlays`	`335190a4`	clean	retain as chart source
`moar_ai_skillz`	`4880f0d3`	clean	reject generated deployment state; remove after archival

The root worktree also has user-owned .gitignore changes and untracked .modden/, todo/, and wkts/. No cleanup operation may touch them.

Stale Branch Cleanup

Deletion is a final post-landing operation, not part of extraction. Preserve a mapping from each removed ref to its replacement merge commit.

Already Merged or Fully Superseded

Propose removing local and matching remote topic refs for:

brokers_refinery*
hist_backfill_fixes* backups
CONT2_hist_fixes
cont_hist_fixes
ib_venue_closures*
stop_is_oec
stop_is_eoc
providers_sync
port_to_latest_tractor
to_asyncio_api_update
tractor_struct_and_godw_mod
tpt_closed_and_finally_footguns*
fix_tractor_logging_POSTREBASEBACKUP
wayland_nix_py313
nixround2*
bump_polars and dev/bump_polars
pdbp_bump
ems_no_last_required
decimal_prices_thru_ems* backups
binance_api_3.1*
alt_tpts_for_perf*
kucoin_and_binance_BACKUP

Remove After Replacement PRs Land

dpi-font-auto-calc*
dpi_scaling_round2
dpi_font_autocalc_fomo_bits
axes_w_zoom_refinery
gap_annotator_BACKUP
dev/gap_annotator
service_mng_to_tractor_PRE_MAIN_REBASE
local aggregate service_mng_to_tractor
repair_tests
multiaddrs
datad_post_mortem duplicate
dev/wkt/fix_broadcast_consumers duplicate
backfiller_deep_fixes / ib_methproxy_refinery aliases
chart_local_gapper aliases
flake_update aliases
worktree-ngi0_submission
py_codestyling aggregate
moar_ai_skillz local/dev/gitea refs

Do not blanket-delete macos_fixes_*, macos_*with_readme, macos_dnks_vibed_fixes, Deribit sources, or nested_order_mode_block until their unrelated unique commits are preserved or explicitly rejected.

Abandoned Object Sweep

No additional recovery lane is required.

Unreachable datad and backfiller commits are patch-equivalent to retained canonical lines.
Temporary chart commit-plan trees match final chart commits.
Unreachable multi-address and order-mode experiments are superseded or contain debugging pauses/cancellation suppression.
Seven stashes contain local dependencies, lock churn, skill links, or .gitignore work.
Stash 4500bb12 contains a unique profiler predicate change, but it makes the default environment mode ineffective and should not be recovered.

Keep stashes untouched until their owners explicitly dispose of them.

Completion Criteria

The repository is ready for branch cleanup only when:

every selected lane has a clean main-based branch and verified forge head/base OIDs;
extracted product trees have been compared against their source intent;
dependency pins are reproducible and contain no editable local paths;
lane-specific tests and the full supported suite pass;
accepted PR merge commits are recorded against all source refs;
no dirty worktree or stash is touched by cleanup;
deferred branches retain explicit owners and reasons for preservation.

18 KiB Raw Blame History