Repository navigation
Conversation
GE3's four-frame series Run reached the page in one 125 ms drain together with its terminal, so no frame was drawn while the Run went and a ring context keyed on the frame survived (probes p08, p09). The row now drives a live Int 2D Run on a watched directory: each frame is written only after the previous one is drawn with its own image and LaB6's rings while Stop is still enabled, and the rings are listed once, before and after Stop. The frames carry no metadata sidecar: a sidecar read that overlaps a new arrival fails the live Run with "metadata discovery changed during read" (pre-existing at 305df40, ledgered separately as LIVE-METADATA-DISCOVERY-RACE). Test only; the series-Run helper is replaced by the live one. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
Tests only, before any production change. HR1: a .txt read during which another frame lands beside the image returns the image's own sidecar. HR2 and HR3 (preservation): a candidate created during an Auto read is still refused at once; a directory that changes during every read is still refused after a bounded number of reads. GR1: a real live Run keeps going when the next frame lands inside a metadata read on the Run's own thread. Each change is a real file created inside the reader's discovery window. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…he fix) The frozen HR2 asserted that a candidate sidecar created during an Auto read is refused at once. Auto records only the candidates that exist when a read begins (probe p02), so a created candidate is visible only as a directory change, exactly like an unrelated arrival. HR2 now pins what the guard can tell apart and keeps refusing at once: a recorded candidate that is rewritten during the read. Tests only; the production fix is not yet committed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
LIVE-METADATA-DISCOVERY-RACE, pre-existing at 305df40: a live directory Run whose images carry metadata sidecars stopped with "metadata discovery changed during read: <directory>" whenever an image arrived while the Run read an earlier frame's metadata. The reader records the image's directory (the true dependency of a case-insensitive sidecar lookup) and refused any read during which it changed; every arrival changes it. - MetadataDiscoveryChanged(OSError): raised, with the unchanged message, when the only discovery inputs that changed are directories that stayed directories. - _with_discovery_revisions classifies every change instead of raising at the first: directory-only changes raise the typed error, any other change keeps the plain OSError. - read_image_metadata_observed reads again after the typed error, five reads in all with 5, 10, 20 and 40 ms pauses, and returns the first read whose inputs held still, with that read's own recorded revisions; the fifth refusal propagates. Still refused at once: a recorded candidate that changes, a recorded directory replaced or removed, and the selected source changing during the read (test_p36_average's transient-foreign-bytes row pins it). A sidecar created mid-read was never recorded as a candidate, so its creation is a directory change and is read again. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…GS2) Tests only, before the M2 production change. A real live Run's validation is held at its entry until Stop is pressed, then runs for real: GS1 holds the loop's re-check of a settled group, GS2 the construct-time check of a newly arrived frame. Both expect a STOPPED terminal. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
GS1 and GS2 held the live loop's validation at its entry, where the alias
check already raises "admission cancelled" and the Run ends STOPPED, so both
passed on their parents (probe p05). The defect lies past that check: a Stop
during the per-image qualification surfaces as InterruptedError("source
qualification cancelled") and the Run ends FAILED. Both rows now hold the
first metadata read inside the selected validation. GS1 (settled re-check)
is the fail-before row; GS2 (construct-time check) is a preservation row,
since _construct's own re-validation already turns it into a stop. Tests
only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
LIVE-STOP-DURING-REVALIDATION, pre-existing at 305df40: pressing Stop while the live loop re-validated a settled group could end the Run FAILED ("source qualification cancelled") instead of STOPPED. Past its alias check (which raises "admission cancelled"), the execution graph's per-image qualification signals cancellation with InterruptedError, and validate_planned_source passed it through, so the executor's top level, which treats "admission cancelled" as a stop, recorded a failure. validate_planned_source now translates that InterruptedError into RuntimeError("admission cancelled"), exactly as _qualify_for_admission does for admission. Every other exception keeps its path. _construct's existing cancellation handler now receives the token it expects at once. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
Test-only. Headless rows at the qualification boundary and real live Runs that keep every drained event and read the durable output after Stop. At 1cfccd7, C1, C2, C5, Q1-Q5 and Q9 fail behaviourally; C3, C4, Q7, P1 and P2 pass (discriminators and preservation). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…Q4c (before any production edit) The maintainer chose that a live Auto frame waits one poll, so a sidecar landing just after its image is read. Q3e and Q4c fail at 1cfccd7 (same-extent); Q3f passes there and guards that absence still means no metadata. _published can now wait for a given member count; rows that call it without one are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
… output (LIVE-QUALIFY-REFUSAL) A metadata file that changes while qualification reads it is a typed source-revision change (MetadataSourceChanged in the reader, MetadataRevisionChanged at _qualify_tiff), so a live Run reads it again at its next poll; other OSErrors, parse failures and oversized inputs stay terminal, and Average's graph requalification reports it as AVERAGE_SOURCE_DRIFT. A live frame waits until its sidecar holds still across a poll: an explicit sidecar must exist, and under Auto a frame waits one poll (the maintainer's choice). When a published frame's metadata changes or appears, at any extent, that output stops growing for the Run: published frames, metadata and lineage are kept, later frames are counted skipped, and the reason stays in every later status. O07's alias row now expects the one Auto readiness wait first. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…uisition keeps publishing) Test-only, added after the first implementation (1928391) from self-review: with whole-group readiness, frames arriving faster than the 0.1 s poll held the whole scan back until the acquisition paused, which 1cfccd7 does not do. P3 guards that each frame waits only for its own metadata. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…ng the whole group (LIVE-QUALIFY-REFUSAL) Found in self-review before handback (P3): with whole-group readiness, frames arriving faster than the 0.1 s poll held a scan back until the acquisition paused. Admission now takes the leading members whose metadata is ready; a later ready member still waits behind an earlier pending one. A partial attempt publishes its prefix but does not settle the group, so the waiting frame is retried next poll, and an attempt that adds nothing beyond the published frames is skipped. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
S1-S9 (headless: SPEC growth keeps an admitted graph valid and undrifted, a rewrite does not, one SPEC state per qualification, nothing new persisted, no grouping inference while live) and L1-L8 (live Runs: row after or before its image, an image never reduced with another point's row, a rewritten published row, a later scan, Auto beside a SPEC file, a fast acquisition, a missing SPEC file). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…EC extent A live frame under SPEC metadata waits for its own complete SPEC row; a live Run never infers images per point. An ordinary SPEC TIFF qualification reads its scan once (read_spec_points) and captures the SPEC file once; each member carries, in memory only, the extent of SPEC bytes it was read from. Alias, member, deferred-state and published-drift checks accept a SPEC file that has only grown past that extent; a rewrite still changes it. Persisted records, the execution identity and Append digests are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…n Append (test-only) Self-review after the first implementation: an image whose SPEC row never comes keeps its group re-admitting at the published extent, and once a later scan grows the SPEC file, Append mode refuses that admission as "same-extent source identity changed" and the Run fails. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…al stitch2d) Composes the paused first implementation with Codex's ratified B1-B3 fixes before the kickoff's CR1-CR5 dispositions. No conflicts; no edit beyond the merge. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…ter review (test-only) C1-C4 (headless): a live qualification waits for a missing row with a typed SourceRevisionChanged and never borrows one; a complete line that is not a row is refused with a ValueError naming it; published rows that are gone or invalid are admitted without metadata so the published- metadata fence reports them; a SPEC file truncated during the read leaves the process alive. R1-R10 (live Runs): a row removed after readiness, a malformed new row, a row rewritten before the reduction read; the SPEC file removed, a covered row or header rewritten, and a raw rewrite with its row while a frame waits; Stop inside the currency check and inside JIT revalidation; progressive publication; a saved live output replays and Appends unchanged inputs and refuses both after the SPEC file grows. No production change; the frozen S, L and P1 files are not edited. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…rrency, a bytes SPEC read Codex's pre-writer review (CR1, CR2, CR6), on the composed cc5af3d: - Live qualification and its re-checks hold each member to its own valid SPEC row (a complete line of the declared column count of numbers, read the same way by silx) and never fall back to the finite per-image reader. A published member whose row is gone or invalid makes every member come back without metadata, for the published-metadata fence; a new member whose row is not there yet is the existing typed retry; a complete line that is not a row at or before it is refused, naming the file, scan and row. - The admitted row values stay in memory; a live Run's reduction read must consume exactly them, and a metadata file changing under that read is retried next poll instead of failing the Run. - SPEC readiness exempts the frames the Run already published (B1's rule). - When nothing beyond the published frames is ready while a frame waits, the attempt waits without admission while the published graph still validates exactly; Stop inside that check stops the Run; any other outcome is admitted as before (B1, B2 unchanged). - The scan read copies the file once instead of mapping it, so a SPEC file truncated or rewritten during the read cannot fault the process. Finite reads are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…uments through (test-only) test_dependency_graph_state_cannot_rebaseline_after_jit_freeze replaces _directory_items with a double that copied its signature; the candidate's new published_members keyword made it a TypeError in the terminal battery. The double now forwards keyword arguments; its assertions are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01D7yaNDFtysn6idFk6L3AAu
…YCLE) The six scattering calibration files closed their CalibrationDialogs and Stitch/RSM pages but never retired or deleted them. One process therefore kept 203,471 widgets / 14,688 top-levels / 3.2 GiB alive after their 169 tests (diagnostic D2 at e52045c). Per-test DeferredDelete delivery cannot delete what was never scheduled for deletion. This extends the maintainer's D5 option 3 (6d61f0d). Each test deletes what it built once it is clean, through a small per-file _dispose. The helper polls the owner's own cleanup receipt to CLEAN, bounded at 30 s, asserts CLEAN with the receipt as diagnostic, then calls deleteLater(); the boundary delivers. - Standalone dialogs use dialog.shutdown(). - Page-mounted dialogs are retired by their page: StitchPage.shutdown(), or RSMPage.close_page() dispatched explicitly in turned_detector, whose _window helper now returns the page. The page deletion takes its window along. - Existing close gestures, barrier releases and body assertions are kept, including guided's immediate-CLEAN assertion. - Rows that already delete their windows (the QMainWindow Escape row and two guided rows) are unchanged. - Guided rows that had no try/finally get one around their unchanged body. Also included, under the pre-writer's option (i): the stale bounds pin goes from 6 to exactly 7. Since 6a7e9f1 a session that records resume is version 7. Codex pre-writer N-f0270e27: PASS WITH AMENDMENTS. Design N-25a21f03: A. QT-TEST-LIFECYCLE Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019FMHuUzRBTwyehRCsUNBxp
CalibrationDialog._restore_session_when_ready re-arms a 20 ms QTimer.singleShot while a resumed session waits for metadata, a worker or the beam probe. The timer had no context object. shutdown() and close() clear the pending resume, so a later firing returns immediately. But a timer already armed kept its callable, and with it the dialog wrapper and the parentless widgets that wrapper holds, alive past the window's deletion until it fired. Diagnostic P5g: the bounds session row (version-7 resume) left 477 widgets / 141 top-levels at its boundary without this change and 0/0 with it, with no added wait. Giving the timer the dialog as its context makes Qt cancel delivery when the window is destroyed. The 20 ms delay, readiness conditions, resume identity fence and restoration body are unchanged. Pre-existing since eb6c94a (2026-10-15). Codex design N-25a21f03: A. QT-TEST-LIFECYCLE Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019FMHuUzRBTwyehRCsUNBxp
HOSTED-REDS-20261017 test-only families (Codex pre-writer N-ccf76dd2): - F2: the P3 ProbeSource.to_scan double takes the executor's cancelled=. - F4: the lattice fraction-reason contract splits: its store half stays in tests/core (no GUI import); its two shared-table blocks move verbatim to test_results_table.py, reusing the core helpers. - F6: the threshold session pin is version 7, as for bounds. - F7: the guided rough-geometry and geometry rows retire the dialog through the bounded _close receipt retry, as the frame-table row does. - F8: P1 records each guideline bar, the failures and whether bars were enforced; the final bar assertion is enforced only with CI absent (maintainer: report-only on hosted CI). Every functional assertion, BARS value and quantile is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019FMHuUzRBTwyehRCsUNBxp
HOSTED-REDS-20261017 F1 frozen regressions on the real projection builder (red at the parent for the four discriminating cases): - a native 2-theta pattern with Share on keeps 2-theta unless Int 2D draws a complete cake, whose Q it then shares; - a nonresident derived GI map (no cake drawn, display retained) keeps its native q_ip trace; resident, Share follows the derived cake. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019FMHuUzRBTwyehRCsUNBxp
Stitch keeps ScientificPreferences.share_axis on for every result since c1dbc3a, and the shared builder then forced the plot axis onto the image axis even when no cake was drawn (Stitch 1D, a thumbnail-only or nonresident frame), overriding an explicit 2-theta, a native 2-theta result or a GI cut. Sharing now applies only in Int 2D with a complete cake, as ScientificView already links it (HOSTED-REDS-20261017 F1). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019FMHuUzRBTwyehRCsUNBxp
HOSTED-REDS-20261017 F1 test completion (Codex F1_AMENDMENT, option a): a shared Stitch 2D map holds the plot axis (LF-011's default; the plot selector is disabled while linked and Share is reached through Options, LF-013), so the GI-cut, chi and read-back rows turn Share off before their first independent selection, as a user must. Every science assertion is unchanged. The positive Options row now also pins the plot selector disabled while shared, enabled after Off and disabled after On. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019FMHuUzRBTwyehRCsUNBxp
HOSTED-REDS-20261017 F3 frozen regressions, driven by real Qt focus and the real metadata worker (red at the parent for both roles): focus leaving a calibrant or direct-beam source field its role's last read already used starts no read; a changed value, an edit undone before leaving, Return and choosing the same file again still read, and one role's read never stands for the other's. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019FMHuUzRBTwyehRCsUNBxp
A calibration source field's editingFinished started a metadata read on every focus-out, so leaving an unchanged field (a checklist row moving focus, a tab, another window) re-read the source and made the window busy (HOSTED-REDS-20261017 F3; exposed by b8c3444's re-activation). Each role now records the inputs of its last applied read; any edit to them or a new read clears it. A commit reads on Return or on inputs other than those, so a change, an edit undone before leaving and both roles stay independent, and the file picker, Read metadata and pending requests still read through their own routes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019FMHuUzRBTwyehRCsUNBxp
…t-only) The eight W18/W40, W20 and W23-regular rows fail only on hosted Linux. Print the reservation, closed and replacement (dev, ino, mode, size, mtime, ctime) tuples, the mount type, the refusal and the facts publish returns, without changing any assertion of the original unlink/recreate rows. W49 and W50 add controls beside them: a held descriptor keeps the original inode allocated, so the replacement cannot reuse its identity. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01467C8Z5X76RUS1xk16SZBe
Codex's pre-push review counts the /proc/mounts reader as evidence machinery. The filesystem type is recorded as unmeasured; every inode/state, refusal, post-abort and hook receipt, every original assertion and W49/W50 stay. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01467C8Z5X76RUS1xk16SZBe
W51 replaces the candidate as soon as reservation returns and requires the session's and publisher's first use to refuse it, preserve the foreign bytes and name no orphan. W52 proves the real exclusive creation descriptor stays open across the substitution (original dev/inode, st_nlink 0) and closes once at retirement. W53 covers retirement on abort, on a publisher writer failure and after a faulted h5py close, with a really-closed pin reported ambiguous once. The close-fault test's reservation arm now binds to the creation descriptor on POSIX; Windows keeps its immediate reservation close. Parent polarity at 53673dc on macOS: W52 and W53 fail (the creation descriptor is closed during reservation), the reservation arm fails, W51 and the other close-fault arms pass. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01467C8Z5X76RUS1xk16SZBe
Descriptor numbers recur. In W53-writer the publisher's source check closed descriptor 13 just before reservation created the pin with that number, so closes.count(pin) saw 2. The spy now starts its close record when it records the creation descriptor. Parent polarity at 53673dc is unchanged: W52/W53 and the reservation arm fail for the same reasons, W51 passes on macOS. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01467C8Z5X76RUS1xk16SZBe
Hosted Linux handed a freed inode number to the next file created, so a candidate replaced by unlink-and-recreate carried the reservation's (st_dev, st_ino). Cleanup, fsync and final-name observation compare only that pair: the foreign file was deleted, named as our orphan, or published as a successful terminal. On POSIX both reservation helpers now return the exclusive creation descriptor still open. It keeps the inode allocated, so no other file can carry the pair while the guards compare it. A failed admission cleans up while pinned and then closes once. The session holds the pin from reserve (or adopt) to _release; the publisher holds it through writing, validation and cleanup and closes it in close_parent, or hands it to adopt. A close fault is reported once. Windows keeps its immediate close (a held CRT handle would block the final rename) and returns no pin. The identity comparisons themselves are unchanged. Pre-existing: the identity-only guards date from 70012f9 and 8d16052. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01467C8Z5X76RUS1xk16SZBe
…nds (test-only) The two hosted reds of test_saved_stitch_2d_navigation_keeps_the_requested_curves were the folder catalog landing between reading the rows and a footer step. The original rows now read their paths after the catalog lands. New rows hold the real folder read on an Event: ResultDirectoryBrowser lists a not-yet-catalogued result where the catalog will (exact-name tie-break, ".." first, no GUI-thread stat; arrival order with no folder); a Stitch run's rows, footer step and no-selection fit batch keep the held order after the catalog lands; RSM's rows and lineout batch likewise. The six RSM stand-ins now list in folder order. Parent polarity at 77e1b9c (macOS): the owner row, both Stitch rows, the RSM pin and the RSM batch row fail on their ordering assertions after the gate entered; the two original navigation rows pass. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01467C8Z5X76RUS1xk16SZBe
…election batch (test-only) The run leaves its result selected, so _fit_batch batches that selection, as designed. The gated Stitch row now clears the selection before reading the held rows and the no-selection batch. Parent polarity at 77e1b9c is unchanged: the row fails on the row-order assertion after the catalog lands. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01467C8Z5X76RUS1xk16SZBe
After a run, Stitch and RSM listed new results in arrival order until the background folder read landed and re-sorted the rows by name, so a row could move under a click and the footer's previous/next stepped through a different order than the one just shown (both hosted reds of the Stitch 2D navigation test). ResultDirectoryBrowser.entries now sorts an active folder's rows with the catalog's own key, browser_catalog.catalog_order_key: ".." first, natural name, then the exact name (scan02 and scan2 no longer fall back on read or arrival order). With no folder, rows keep arrival order. No filesystem call on the GUI thread. Pre-existing: the ordering rule is unchanged since before 9e82766. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01467C8Z5X76RUS1xk16SZBe
Codex's ORACLE_RULING (145f65fb): these rows picked results by row position assuming stitched order, which held only until the folder catalog re-read. With rows in folder order from the start, the shown-only residency row binds first, missing and latest to the page's stitched order and keeps the exact visible tuple; the pin row selects a listed result other than the pinned one; the GI/plain row uses the GI run's own target and the plain family; the frame inspector switches to a listed result other than the current one; the RSM row revisits two, three, two by identity. Every original behavior assertion stays. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01467C8Z5X76RUS1xk16SZBe
"Finite-result saves on POSIX preserve file identity..." now says what users see: on Linux and macOS, Stitch, RSM, Average and Reintegrate saves no longer delete or publish a file another process put in their place. "First-seen files still require confirmation" read like a dialog; a newly seen file is processed once a second directory listing confirms it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01467C8Z5X76RUS1xk16SZBe
…nly) Codex's terminal review of STITCH-BROWSE-ORDER-20261017 recorded two non-blocking test-completion residuals: - The gated Stitch and RSM rows now keep a real selection across the catalog landing and assert the current result, the selection, resident key/payload identity and the selected batch survive it; the no-selection batch follows the rows; a multi-result selection batches in Qt's selected order. - The rows assert the catalog actually landed with the held results, and _held_catalog raises if a held read is never released instead of reading late. Candidate: the three files 20 passed. Pre-fix product (77e1b9c src): the gated rows still fail at the row-order assertion, after the landing check. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01467C8Z5X76RUS1xk16SZBe
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Draft, not for merge. This PR exists only to run
pr.yml's hosted jobs on the integrated testing branchstitch2d(0dfaa4e3).The run of interest is
windows-pixi-native, whose new step runstests/core/test_windows_finite_publication.pyon GitHub's Windows runner (local NTFS). It covers Average, Stitch and RSM finite publication without directory descriptors. It does not replace the maintainer's own Windows NTFS and SMB qualification.🤖 Generated with Claude Code
https://claude.ai/code/session_01L1UPY7Ep5ASRCWzUByxzNf