Roll out new toolchain support in thin vertical slices with fixture-backed coverage

Summary#

The strongest multi-PR pattern for feature expansion was to avoid a one-shot integration and instead land discovery, execution, hashing, watch behavior, pruning, and CI support as separate layers. This keeps each slice reviewable, preserves existing behavior, and makes it easier to attach representative fixtures and platform-specific tests at each step.

Problem#

Adding a new workspace or platform model touches graph construction, task execution, invalidation, cache identity, pruning, and developer ergonomics. Bundling all of that at once makes regressions hard to isolate and usually leaves at least one lifecycle path under-tested.

Solution#

Introduce a foundational composition/refactor layer first, then land capability slices one by one with explicit non-goals, realistic fixtures, and end-to-end tests for mixed and pure environments. Fail closed on unsupported or unsafe states, and pin CI/runtime dependencies so coverage reflects actual supported behavior.

Failure Modes#

Shipping discovery without hashing/watch/prune semantics, leading to stale or incorrect builds
Testing only happy-path fixtures and missing mixed-workspace edge cases
Letting optional-toolchain support leak into existing code paths before a shared composition layer exists

Sources#

https://github.com/tensorflow/tensorflow/pull/124430
https://github.com/tensorflow/tensorflow/pull/124472
https://github.com/tensorflow/tensorflow/pull/124462
https://github.com/tensorflow/tensorflow/pull/124398
https://github.com/tensorflow/tensorflow/pull/124414
https://github.com/tensorflow/tensorflow/pull/124458
https://github.com/tensorflow/tensorflow/pull/124419
https://github.com/tensorflow/tensorflow/pull/124396
https://github.com/tensorflow/tensorflow/pull/124478
https://github.com/tensorflow/tensorflow/pull/124479
https://github.com/tensorflow/tensorflow/pull/124390
https://github.com/huggingface/transformers/pull/47682
https://github.com/huggingface/transformers/pull/47681
https://github.com/huggingface/transformers/pull/47680
https://github.com/huggingface/transformers/pull/47679
https://github.com/huggingface/transformers/pull/47587
https://github.com/huggingface/transformers/pull/47663
https://github.com/huggingface/transformers/pull/47673
https://github.com/huggingface/transformers/pull/47650
https://github.com/huggingface/transformers/pull/47322
https://github.com/huggingface/transformers/pull/47652
https://github.com/microsoft/ML-For-Beginners/pull/1002
https://github.com/microsoft/ML-For-Beginners/pull/1001
https://github.com/microsoft/ML-For-Beginners/pull/1000
https://github.com/microsoft/ML-For-Beginners/pull/994
https://github.com/microsoft/ML-For-Beginners/pull/991
https://github.com/microsoft/ML-For-Beginners/pull/990
https://github.com/microsoft/ML-For-Beginners/pull/989
https://github.com/ClickHouse/ClickHouse/pull/111762
https://github.com/ClickHouse/ClickHouse/pull/112603
https://github.com/ClickHouse/ClickHouse/pull/110451
https://github.com/ClickHouse/ClickHouse/pull/112524
https://github.com/ClickHouse/ClickHouse/pull/108326
https://github.com/ClickHouse/ClickHouse/pull/112830
https://github.com/ClickHouse/ClickHouse/pull/111060
https://github.com/ClickHouse/ClickHouse/pull/112810
https://github.com/ClickHouse/ClickHouse/pull/112527
https://github.com/ClickHouse/ClickHouse/pull/112670
https://github.com/ClickHouse/ClickHouse/pull/110606
https://github.com/ClickHouse/ClickHouse/pull/112223
https://github.com/ClickHouse/ClickHouse/pull/108977
https://github.com/ClickHouse/ClickHouse/pull/109232
https://github.com/ClickHouse/ClickHouse/pull/112820
https://github.com/vercel/turborepo/pull/13623
https://github.com/vercel/turborepo/pull/13522
https://github.com/vercel/turborepo/pull/13602
https://github.com/vercel/turborepo/pull/13622
https://github.com/vercel/turborepo/pull/13613
https://github.com/vercel/turborepo/pull/13612
https://github.com/vercel/turborepo/pull/13621
https://github.com/vercel/turborepo/pull/13611
https://github.com/vercel/turborepo/pull/13610
https://github.com/vercel/turborepo/pull/13609
https://github.com/vercel/turborepo/pull/13616
https://github.com/vercel/turborepo/pull/13608
mined_at: 2026-08-01T03:36:33Z

Sagwan Revalidation 2026-08-01T04:16:01Z#

verdict: ok
note: 일반적 롤아웃·테스트 원칙으로 현재 practice와 충돌 없음

Sagwan Revalidation 2026-08-03T16:07:48Z#

verdict: ok
note: 최근 검증 이후 변동 가능성 낮고, 얇은 수직 슬라이스 권장도 여전히 유효함

Sagwan Revalidation 2026-08-07T13:16:38Z#

verdict: ok
note: [chatgpt HTTP 401] {

Sagwan Revalidation 2026-08-10T00:22:09Z#

verdict: ok
note: [chatgpt HTTP 401] {

Sagwan Revalidation 2026-08-12T11:59:06Z#

verdict: ok
note: [chatgpt HTTP 401] {

Sagwan Revalidation 2026-08-15T01:23:25Z#

verdict: ok
note: [chatgpt HTTP 401] {

Sagwan Revalidation 2026-08-17T13:17:04Z#

verdict: ok
note: [chatgpt HTTP 401] {

Sagwan Revalidation 2026-08-20T01:32:23Z#

verdict: ok
note: [chatgpt HTTP 401] {

Sagwan Revalidation 2026-08-22T14:07:09Z#

verdict: ok
note: [chatgpt HTTP 401] {

Sagwan Revalidation 2026-08-25T02:11:30Z#

verdict: ok
note: [chatgpt HTTP 401] {

Sagwan Revalidation 2026-08-27T14:29:01Z#

verdict: ok
note: [chatgpt HTTP 401] {

Sagwan Revalidation 2026-08-30T03:03:02Z#

verdict: ok
note: [chatgpt HTTP 401] {

Sagwan Revalidation 2026-09-01T16:30:55Z#

verdict: ok
note: 일반적 롤아웃 패턴이라 최근 practice와 충돌 없고 재사용 가능함

Sagwan Revalidation 2026-09-07T20:58:37Z#

verdict: ok
note: [chatgpt HTTP 404] {

Sagwan Revalidation 2026-09-10T09:27:17Z#

verdict: ok
note: 최근 검증 이후 변동 가능성 낮고 일반적 롤아웃 권장안도 유효함

Sagwan Revalidation 2026-09-13T04:23:53Z#

verdict: ok
note: 박층 슬라이스 롤아웃 + 픽스처 커버리지 패턴은 2026년에도 표준 실천법이며, 주장·실패 모드 모두 여전히 유효하다.