[ScriptedAlchemy]

Merge #4478 first, followed by #4526, #4510, #4524, #4489, #4485, #4527, #4528, #4530, #4531, #4529, #4534, and #4541.

This PR is number 3 of 13, based on #4526. Merge #4524 next.

#4516 has already merged into this branch.

Use merge commits. After merging a parent PR, change its child's base to master before merging the child. Auto-merge is disabled.

A cold build that exceeds the resident-memory limit now waits until enough memory is available for the full reconstruction. Cold publication also avoids holding a decoded repository alongside its encoded output.

The existing memory authority tracks allocations owned by the operation. It retains the full memory requirement after a refused build unwinds. A prepared publication can resume without repeating completed extraction. Worker admission measures current pressure and reserves memory for the installed worker count. Before activation, publication verifies every referenced digest and exact extent, including inherited segments.

Cold builds now reserve memory before replacement allocations and compact copies. They limit temporary carry data to one file, reuse authenticated unchanged segments, release resolution facts before building the graph, and defer projection rows. Symbol-name lookup and resolution-index page construction request memory in batches. Digest serialization streams explicitly ordered objects and preserves the exact bytes.

Fixes #4505.

Validation

Before the final integration of ancestor commits, the following tests passed:

• Runtime-core passed 542 tests.
• Index unit tests passed 357 tests.
• Index integration passed 216 tests.
• Scheduler runtime passed 664 tests, with 2 existing ignored tests.
• The reporting integration regression passed separately.

Additional checks passed 96 retention tests, 4 resident-accounting tests, 75 query integration tests, and 27 evaluator tests. Changed benchmark callers compiled. Scoped Clippy and formatting checks passed.

The ignored scale fixture with 6,600 files and 422,400 symbols published under the unchanged 1 GiB limit in 241.64 seconds. Peak admission occupancy was 805,367,808 bytes. Peak RSS was 850,829,312 bytes. The run did not decode the full generation eagerly.

At stacked head 3e10b5fbb0a5d49b59c7fbe7216e43c6f10b0a2d, with nightly-2026-10-25 and inherited dependencies, fresh execution passed 542 runtime-core, 357 index unit, 223 index integration, and 664 scheduler runtime tests. The 2 existing ignored tests remained ignored. All three affected formatting targets passed. Bazel invocation 27ebabd4-4bd2-4576-83c9-64be264208cc exited with code 0.

The scale result covers cold publication. It does not establish a bound for retained-text serving or every allocator. Existing limits in decoder preallocation and measurement of native, parser, and Go internal workspaces remain. No budget or timeout increased.

Publication cancellation correction

A source epoch changing inside final durable commit surfaced as CompareAndSwap and could park the project. The scheduler now restores cancellation only when the attempt control confirms it, preserving genuine pointer conflicts and memory-refusal precedence. Fix: dab12b4.

Focused regression BEFORE: 0 passed / 1 failed (fa27d3a7-0aa4-4164-a350-6f943e556f4c). AFTER publication-store module: 45 passed plus formatting (836e4ba5-46ab-4200-aa71-af891b3acd38). Full daemon suite: 79 passed, including all three CI failures (711bbf0f-9197-456f-be73-3746d23f04bb). Scoped Clippy passed (bfa5119d-e483-4697-9846-3d10e9ff3330). These ran on the #4510 correction before stack propagation. The separate refresh-memory journey passed with existing phase tracing in 158.23s (a7c28c80-c892-4279-ba35-5d4beed34574), then passed again after removing all investigative edits: 2 tests, 0 failures, 157.87s (fc944727-2cd0-4620-b7bc-86cea6ea5047). Both completed cold indexing and all four refreshes under the unchanged 6 GiB limit and 300-second target timeout. The slower CI execution was not reproduced; no performance fix or timeout increase is claimed.

[changeset-bot]

⚠️ No Changeset found

Latest commit: dab12b4

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR