compass(detect): the set-iteration lint over atom/compass, carried from #96 by jgong5 · Pull Request #512 · jgong5/ATOM

jgong5 · 2026-09-30T06:57:13Z

Closes #487. Ready for review. No blocking issues.

Adds atom/compass/detect/set_iteration.py, carried from #96 at ee5b591dc, and gates atom/compass with it in tests/compass/test_set_iteration_lint.py. Stacked on #511, whose ClockSourceLint.modules, scopes and scope_at it imports.

Dev record

Found: at 0e16d3a98 the carried lint reports the same two sites compass: a determinism harness two processes wide, and the set-iteration lint #96's base update found: SimulatedKVConnector._matured in atom/compass/kv/connector.py and capture_token_shapes in atom/compass/memory/graph_pool.py.
Decided: both sites are settled in code, and the lint carries no exemption. _matured collects finished ids in a list, in the dict's order. capture_token_shapes sorts the set before filtering it, which returns the same tuple. graph_pool.py is outside the issue's file set; the review asked for this change there. A comprehension handed to sorted is reported, because its element and filter run in the set's order.
Surprised: none.
Left undone: atom/utils/clock.py is not a scan root because compass(clock): the LP runtime — TAR, NER, send registration and TSO release inside an ATOM process #455 has not landed; whichever of the two lands second adds it. Both static passes name their own blind spots and both lists are short #110 still holds the recognisers compass: a determinism harness two processes wide, and the set-iteration lint #96's reviews routed there.
Watch item for compass(kv): time the simulated KV transfer on Mooncake's scheduler-side hooks instead of the worker #466: Scheduler._update_from_kv_xfer_finished frees blocks in the order of the sets the connector returns. It is outside the scan root, so this PR does not change it.

Named result

atom/compass goes from 2 reported sites at 0e16d3a98 to 0 at the head, both settled in code.

Gates

branch @ 0554449a3: 5327 passed, 155 skipped, 3 xfailed, 0 failed, GATE_CPU_RC=0. ruff check and format clean on the three changed files, rc 0.
control compass(detect): the clock-source lint over atom/compass, without asyncio.sleep #511 @ 8056b1933: 5238 passed, 155 skipped, 3 xfailed, rc 0. The +89 are this PR's tests.

Evidence

Revert-red: one line-preserving edit of the head tree each, run over tests/compass/test_set_iteration_lint.py (89 pass unedited).

edit	failed	red node
`capture_token_shapes` back to `tuple(sorted(s for s in shapes if ...))`	1	`test_the_simulated_path_is_clean_today`: `assert 1 == 0`
`_sorted_away` and its call reinstated	1	`test_every_form_it_claims_to_report_is_reported[sorted-generator]`: `assert () != ()`
`connector.py` back to `0e16d3a98`	1	`test_the_simulated_path_is_clean_today`: `assert 1 == 0`
`ClockSourceLint.modules` walks the top directory only	1	`test_the_simulated_path_is_clean_today`: `kv/connector.py` not among the scanned modules

The last two rows are from eb29a5a92; neither line changed since. Reinstating the exemption is not line-preserving; no test in the file reads line numbers of set_iteration.py.

capture_token_shapes old and new forms return equal tuples on 2004 inputs: four fixed ladders and 2000 random ones, with empty ladders and a zero budget among them.

Changes to the carried code, relative to ee5b591dc:

set_iteration.py: +9 / -6 lines. The public scopes and scope_at imports, one docstring sentence saying a comprehension under sorted is reported, and a _binds docstring that states no count.
Test file: the set-iteration tables, the reading_sets fixture and TestTheSetIterationLint, with a new module docstring and the walk assertion. It adds the sorted-generator fire form and leaves out the clock-source cases, which are in compass(detect): the clock-source lint over atom/compass, without asyncio.sleep #511.
connector.py: 3 lines changed. graph_pool.py: 1 line changed.

Cost, as docstring-inclusive AST statements / non-blank physical lines at 0554449a3:

set_iteration.py: 278 / 486 (carried 278 / 483).
Test file: 103 / 527.

Generated with Claude Code

…ncio.sleep Carries ClockSourceLint from the closed #96 head (ee5b591) with its anchored read comparison, and its tests from #91 and #96, into atom/compass/detect/clock_source.py and tests/compass/test_clock_source_lint.py. Changes from the carried module: - asyncio.sleep leaves CLOCK_READS: an asyncio timer runs on the event loop's clock, which a simulated run replaces. - The watchdog allow-list entry goes with the watchdog; the list is empty. - The unused enabled flag and scan_tree are dropped. The lint is clean over atom/compass at the tip. Closes #101. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

A static pass that reports every read of a set in hash order under atom/compass: file, line, form and enclosing symbol, parsed with ast. Membership is not reported. Carried from ee5b591, stacked on the clock-source lint (#511) whose module walk and scope lookup it reuses, with two settlements of the sites it reports at the tip: - a comprehension handed straight to sorted() is not a read; it is the same remedy as sorted(s) with a filter (graph_pool.capture_token_shapes); - SimulatedKVConnector._matured collects the finished requests in a list in the dict's own order, so no set is iterated. The tests move to tests/compass/test_set_iteration_lint.py; the clock-source cases they shared are in #511. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

…old recursion, alias chains and innermost scope check() refuses a root holding no module, so a mistyped path no longer reports clean over zero modules. The seeded read sits one directory down, so the walk's recursion is held; the clean-line test checks a single file, so the file-root branch that #512 relies on is held too. CLOCK_READS_THAT_FIRE pairs each spelling with its exact report line, which folds in the three exact-report tests and the datetime test and adds alias-of-alias and innermost-scope cases. _scopes and _scope_at become scopes and scope_at for the set-iteration lint. Dropped: the believable-number test (it ran no repository code), the CLOCKS_HANDED_AROUND table (it asserted blind spots), main and argparse (no caller), scan_modules (one caller, inlined). The alias-expansion record is a set. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

jgong5 · 2026-09-30T07:15:44Z

+        yield from _scoped(child, inner)
+
+
+def _sorted_away(tree) -> set:


REQUEST CHANGES, required 1. A call in the element or filter still runs in the set's order and is not reported: sorted(ids.take(x) for x in s) gives (), the same body as a for loop is reported. Delete _sorted_away and write capture_token_shapes as tuple(s for s in sorted(shapes) if s <= max_num_batched_tokens); atom/compass is then clean. Or exempt only an element and filters with no Call, NamedExpr, Await or Yield, with a fire form.

jgong5 · 2026-09-30T07:15:46Z

+SIMULATED_PATH = os.path.dirname(atom.compass.__file__)
+
+#: A step loop that reads its participants out of a set every way the pass
+#: claims to see, and seven more it does not. Everything here runs; nothing


REQUEST CHANGES, required 2. seven more counts BEYOND_THE_PARSER and goes stale when an entry moves, as do Two false positives (module docstring) and Half of these (above FORMS_THAT_STAY_QUIET). Drop the numbers.

jgong5 · 2026-09-30T07:15:47Z

+        code, report = lint.check(SIMULATED_PATH)
+        assert code == 0
+        assert report == (
+            f"set-iteration lint: clean over {len(lint.modules(SIMULATED_PATH))} "


No blocking issues; ponytail. The second lint.modules(SIMULATED_PATH) walks the tree again. Use len(scanned).

jgong5 · 2026-09-30T07:15:48Z

Review, round 1. Verdict: CHANGES NEEDED. REQUEST CHANGES: 2 blocking, head eb29a5a92, own commit only (7d59d13b0..eb29a5a92).

Required:

_sorted_away hides a call in the element or filter, which still runs in the set's order (inline). It buys one site; sorting the set first in capture_token_shapes instead is clean and about 27 lines shorter.
Three item counts in the test file's prose (inline).

Checked: carried code matches ee5b591dc outside the listed hunks; nothing from the do-not-carry list; #96's lint findings sit in #110, as #487 says; no design-doc references; ruff clean.

Accepted with reservation: the _matured change serves the lint only. Deletion order leaves the same dict, and the issue allows settling a site in code.

Watch next: Scheduler._update_from_kv_xfer_finished iterates the returned sets and deallocates in that order, outside the scan root, while the connector docstring says nothing may turn on it (#466). sorted({x for x in s}) is still reported. #110's exit names tests/compass/clock/test_static_checks.py, now tests/compass/test_set_iteration_lint.py.

branch @ eb29a5a92: 5337 passed, 155 skipped, 3 xfailed, GATE_CPU_RC=0.
control compass(detect): the clock-source lint over atom/compass, without asyncio.sleep #511 @ 7d59d13b0: 5245 passed, 155 skipped, 3 xfailed, GATE_CPU_RC=0.

Evidence

Probes at the head (scan_source, then with _sorted_away returning an empty set):

source	head	no exemption
`sorted(ids.take(x) for x in s)`	`()`	comprehension over s
`sorted([x for x in s if log.append(x) is None])`	`()`	comprehension over s
`sorted((next(counter), x) for x in s)`	`()`	comprehension over s
same body as a `for` loop appending `ids.take(x)`	for-loop over s	for-loop over s
`sorted(x for x in s if x <= m)`	`()`	comprehension over s

Over atom/: 35 reads with the exemption, 36 without; the one silenced is capture_token_shapes. With no exemption and that line written as tuple(s for s in sorted(shapes) if s <= max_num_batched_tokens), atom/compass gives no reads, and the two forms return equal tuples on three inputs.

Revert-red on tests/compass/test_set_iteration_lint.py (92 pass unedited, null-comment control 92 pass), one line-preserving edit each:

edit	result	red node
`_matured` back to `7d59d13b0`	1 failed / 91	`test_the_simulated_path_is_clean_today`
exemption off	3 failed / 89	the same, plus `[sorted-generator]`, `[sorted-list-comprehension]`
`sorted` by attribute tail	1 failed / 91	`[method-named-sorted]`
every comprehension under the argument	1 failed / 91	`[comprehension-inside-a-sorted-element]`
`ListComp` dropped	1 failed / 91	`[sorted-list-comprehension]`

Ponytail review: the exemption and its four test forms can go (above); len(lint.modules(SIMULATED_PATH)) can be len(scanned). net: -27 lines possible.

Generated with Claude Code

Base update. #511 publishes `scopes` and `scope_at` (formerly `_scopes` and `_scope_at`, same signatures) and deletes `ClockSourceLint.scan_modules` and `main`. `set_iteration.py` now imports the public names, and the local that held the spans is renamed `enclosing` so it does not shadow `scopes`. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

…t first in capture_token_shapes A comprehension's element and filter run in the set's order before `sorted` sees them, so `sorted(ids.take(x) for x in s)` is an order dependence the exemption hid. The exemption and its test forms are gone; the fire list carries that case instead. `capture_token_shapes` sorts the set before filtering, which returns the same tuple and reads no set in hash order. The test file's prose no longer states item counts, and the clean-today assertion reuses the scanned module list instead of walking the tree twice. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

jgong5 · 2026-09-30T07:29:34Z

Round 2 pushed 0554449. Both required findings fixed; the ponytail comment taken.

7911684be merges #511 at 8056b1933 and imports scopes and scope_at; the local is now enclosing. 0554449a3 deletes _sorted_away, and capture_token_shapes sorts the set before filtering. The Scheduler._update_from_kv_xfer_finished watch item is recorded in the body for #466.

branch @ 0554449a3: 5327 passed, 155 skipped, 3 xfailed, GATE_CPU_RC=0; control #511 @ 8056b1933 5238 passed, the +89 are this test file.

Findings, fixes and reds

Each defect reinstated on the head tree, over tests/compass/test_set_iteration_lint.py (89 pass unedited, 92 at the merge commit before the form changes).

finding	fix	reinstated defect reddens
1 exemption hides calls	`_sorted_away` deleted with its two quiet forms and two fire forms; `sorted(ids.take(x) for x in s)` added as fire form `sorted-generator`	exemption back: `test_every_form_it_claims_to_report_is_reported[sorted-generator]`, `assert () != ()`
1 `capture_token_shapes`	`tuple(s for s in sorted(shapes) if s <= max_num_batched_tokens)`; equal to the old form on 2004 inputs	old line back: `test_the_simulated_path_is_clean_today`, `assert 1 == 0`
2 item counts	"seven more", "Two false positives" and "Half of these" reworded	none: prose
ponytail `len(scanned)`	the report assertion reuses `scanned`	none: no behaviour change
base update	`set_iteration.py` imports the public names	old import back: collection error, `ImportError: cannot import name '_scope_at'`

ruff 0.16.7 check rc 0, format and black clean on the three changed files.

Generated with Claude Code

jgong5 · 2026-09-30T07:40:18Z

+    "augmented": "def f(n, t):\n    s = set(n)\n    s |= t\n    for x in s:\n        yield x\n",
+    "typing-Set": "from typing import Set\n\n\ndef f(r: Set[str]):\n    return [x for x in r]\n",
+    "string-annotation": 'def f(r: "set[str]"):\n    return [x for x in r]\n',
+    "sorted-generator": "def f(s: set[str], ids):\n    return sorted(ids.take(x) for x in s)\n",


No blocking issue. A list comprehension handed to sorted has no fire form, so only the generator case is held. Put the exemption back for ast.ListComp alone and tests/compass still passes: 1371 passed, 6 skipped. Add one beside this form, e.g. sorted([ids.take(x) for x in s]).

jgong5 · 2026-09-30T07:40:19Z

Review, round 2. Verdict: APPROVE. APPROVE at head 0554449a392538ff1e2197d99b7538f5b10b2c75, delta eb29a5a92..0554449a3.

Both round-1 findings are closed:

_sorted_away is deleted. With it back as it was at eb29a5a92, test_every_form_it_claims_to_report_is_reported[sorted-generator] fails.
The three item counts in the test file's prose are gone.

Checked: the merge 7911684be resolves only the rename to scopes and scope_at. The body records the graph_pool.py change and why it was made. No design-doc references. ruff check and format are clean on the four changed files.

Accepted with reservation: a list comprehension handed to sorted has no fire form (inline).

Watch next: #511 has not landed. The stacked form below takes #511 onto the current tip first. The literal merge-tree --merge-base 8056b1933 91e19c53b <head> deletes clock_source.py, because #511 is not on the tip yet.

stacked (compass(detect): the clock-source lint over atom/compass, without asyncio.sleep #511 then compass(detect): the set-iteration lint over atom/compass, carried from #96 #512 on 91e19c53b) @ tree 920a0630a: 5327 passed, 155 skipped, 3 xfailed, GATE_CPU_RC=0.
control (compass(detect): the clock-source lint over atom/compass, without asyncio.sleep #511 on 91e19c53b) @ tree 3c55a73a3: 5238 passed, 155 skipped, 3 xfailed, GATE_CPU_RC=0. The +89 are this test file.

Evidence

The stacked tree is the head tree plus #478's three files (sync_sites.json, cliargs.rs, worker_manager.rs), none of them in this PR.

Reinstated on the head tree and run over tests/compass/test_set_iteration_lint.py (89 pass unedited):

edit	lines	result	red node, assertion
null control: 17 comment lines where the exemption goes	+17	89 passed	none
exemption as at `eb29a5a92`	+17	1 failed / 88	`[sorted-generator]`, `assert () != ()`
exemption for `ast.GeneratorExp` only	+17	1 failed / 88	`[sorted-generator]`, `assert () != ()`
exemption for `ast.ListComp` only	+17	89 passed; `tests/compass` 1371 passed, 6 skipped	none
`capture_token_shapes` back to `eb29a5a92`	0	1 failed / 88	`test_the_simulated_path_is_clean_today`, `assert 1 == 0`

Ponytail review of the delta: Lean already. Ship.

Generated with Claude Code

…ass/issue-487-set-iteration-lint Base update only. One add/add conflict in tests/compass/test_clock_source_lint.py, resolved by taking the integration tip's version unchanged (the final version of #511; this branch carried an earlier round of it and never changed the file itself). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

jgong5 · 2026-09-30T08:04:57Z

Base update pushed 1be46ea. No blocking issues.

Merged fork/feature/atomcompass_new @ 073cebf (#511 landed). No code change of this PR's own.

Resolved: tests/compass/test_clock_source_lint.py (add/add). Took the integration tip's version unchanged; this branch carried an earlier round of compass(detect): the clock-source lint over atom/compass, without asyncio.sleep #511's file and never edited it.
Merge tree 307d31ba9c55ffcffd39de907655809fceb33afa, identical to the stacked tree gated in compass(detect): the clock-source lint over atom/compass, without asyncio.sleep #511's final review: 5330 passed, rc 0.

jgong5 · 2026-09-30T08:06:41Z

Review, round 3. Verdict: APPROVE. APPROVE at head 1be46ea86a504ccf264fe4580eda55b396a3dc1a, delta 0554449a3..1be46ea86, one base-update merge of 073cebf25 (#511).

Checked: the only remerge hunk is the add/add in tests/compass/test_clock_source_lint.py, taken as the tip's version. On this branch only #511's own commits 7d59d13b0 and 8056b1933 touched that file, so the tip's version (#511 final) drops nothing. 1be46ea86^2..1be46ea86 is exactly this PR's four files. Head tree is 307d31ba9, the tree gated in #511's final review (5330 passed, rc 0), so no new gate run. merge-tree --write-tree against the tip 073cebf25 gives the same tree.
Accepted with reservation: none.
Watch next: none beyond round 2.

Generated with Claude Code

jgong5 and others added 2 commits September 30, 2026 06:36

jgong5 added this pull request to stack #513 September 30, 2026 06:57

jgong5 commented Sep 30, 2026

View reviewed changes

This was referenced Sep 30, 2026

Both static passes name their own blind spots and both lists are short #110

Open

compass(detect): the clock-source lint over atom/compass, without asyncio.sleep #511

Merged

jgong5 and others added 2 commits September 30, 2026 07:19

jgong5 commented Sep 30, 2026

View reviewed changes

jgong5 mentioned this pull request Sep 30, 2026

compass(clock): the LP runtime: TAR, NER, send registration and TSO release #517

Merged

Base automatically changed from compass/issue-488-clock-source-lint to feature/atomcompass_new September 30, 2026 08:03

jgong5 marked this pull request as ready for review September 30, 2026 08:07

jgong5 merged commit d629ddd into feature/atomcompass_new Sep 30, 2026

jgong5 added module: compass-detect atom/compass/detect/ and its tests module: compass-kv atom/compass/kv/ and its tests module: compass-memory atom/compass/memory/ and its tests labels Oct 10, 2026

Navigation Menu

Repository navigation

compass(detect): the set-iteration lint over atom/compass, carried from #96 - #512

compass(detect): the set-iteration lint over atom/compass, carried from #96#512
jgong5 merged 6 commits into
feature/atomcompass_newfrom
compass/issue-487-set-iteration-lint

jgong5 commented Sep 30, 2026 •

edited

Loading

Uh oh!

jgong5 Sep 30, 2026

Uh oh!

jgong5 Sep 30, 2026

Uh oh!

jgong5 Sep 30, 2026

Uh oh!

jgong5 commented Sep 30, 2026

Uh oh!

jgong5 commented Sep 30, 2026

Uh oh!

jgong5 Sep 30, 2026

Uh oh!

jgong5 commented Sep 30, 2026

Uh oh!

jgong5 commented Sep 30, 2026

Uh oh!

jgong5 commented Sep 30, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

		yield from _scoped(child, inner)


		def _sorted_away(tree) -> set:

Conversation

jgong5 commented Sep 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Dev record

Named result

Gates

Uh oh!

jgong5 Sep 30, 2026

Choose a reason for hiding this comment

Uh oh!

jgong5 Sep 30, 2026

Choose a reason for hiding this comment

Uh oh!

jgong5 Sep 30, 2026

Choose a reason for hiding this comment

Uh oh!

jgong5 commented Sep 30, 2026

Uh oh!

jgong5 commented Sep 30, 2026

Uh oh!

jgong5 Sep 30, 2026

Choose a reason for hiding this comment

Uh oh!

jgong5 commented Sep 30, 2026

Uh oh!

jgong5 commented Sep 30, 2026

Uh oh!

jgong5 commented Sep 30, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

jgong5 commented Sep 30, 2026 •

edited

Loading