MagicBeanDev/Longshot

Watch

Fork 98

WIP: fix(build): honor selected external hot-function profiles #754

Draft

siantm wants to merge 1 commit from siantm/Longshot:fix/external-hot-profile into main

siantm commented

2026-11-15 10:00:25 +00:00

First-time contributor

Type: fix · Risk: low · Tested on: Windows 11 Pro 26300, game-free · Needs: complete game-free suite and Linux verification

Why

An explicitly selected hot-function profile can silently lose to the build folder's default list. Editing an external profile at the same path can also reuse a translation produced from its old contents.

How to reproduce

Seen on: Windows 11 Pro 26300, Python 3.14.7; no GPU or game used.
Settings and switches: external or explicitly empty A64_HOT_FUNCS with a different default work/hot_funcs.txt.
Steps: Run the two selected override-consumer tests in tests/test_make_game_cache.py against main using its LONGSHOT_CACHE_TEST_SOURCE override, then against this branch. Run check_selected_hot_list() from tests/test_lift_cache.py against each implementation.
Expected: the chosen list controls the real lifter consumer; empty disables it; editing selected contents invalidates the translation.
Actual: main selects the default list for both overrides and does not invalidate on external-file edits.

What changed

Preserve explicit profile selection, including an empty value, and hash the selected profile's contents. Resolve relative profiles from the translation child's checkout root.

nx2/make_game.py: override precedence and selected-content lift-key inputs.
Tests exercise real environment merging, a64_lift.hot_functions(), cache invalidation, and cross-folder reuse.
Bind the cache fixture's real shipped-profile lookup to its authored checkout and close only its owned Windows lock. The profile-isolation part overlaps the fixture prerequisite in #592.
Scope is distinct from environment-key cleanup in #22 and profile-path object keys in #660.

Measurements

Verdict: the authored regressions fail on main and pass with the fix. No FPS claim.
Hardware: Windows 11 Pro 26300, Intel Core i7-9700, 31.8 GiB usable RAM; pinned Python 3.14.7; no GPU work.
Method: identical authored consumer/cache cases against main and the prepared branch.

Check	Before: main @ `9e6f231d`	After: this PR @ `b6d9ad2af4`
External override and explicitly empty override	Both fail: default function selected	Both pass
Selected file edited at the same path	Cache key incorrectly retained	Lift-key regression passes
Complete cache/consumer driver	Not claimed as a baseline full-driver result	19 pass, 1 platform skip

Observed: FAIL: editing the same external profile path changes the key on main; main's two consumer assertions select f_80000000 rather than the authored external function/empty set. Patched driver: Ran 20 tests ... OK (skipped=1). Lift-key and cross-folder lift-store drivers pass. These checks ran on the source snapshot committed as b6d9ad2af4 before browser submission; no generated game files are included.

Flags and settings

A64_HOT_FUNCS: previously the builder's default list could override an explicit setting; now explicit selection wins, empty disables it, and unset retains existing behavior. No new switch/default or runtime quality change.

Testing

Tested: focused lift-key, complete cache/consumer, and lift-store drivers; authored baseline failure proofs. New run fragment is accepted by the actual tested-configurations renderer.
Not tested: complete suite, native Linux, or an actual game build. The independent tested-configurations unit driver has a Windows CRLF-vs-LF pipe-fixture failure; it is not reported as passing.
Must be tested: complete game-free suite on Windows and Linux, with source commits recorded, before readiness; by hand or the project's Test Farm. No runtime performance claim is proposed.

Risks

What could break: profile selection or cache invalidation could reuse the wrong translated code.
Turning it off: revert the fix or leave the existing override unset.
Compatibility: selected translation inputs can intentionally cause one retranslation. Saves, settings, shader/pipeline cache formats, renderer ABI and tools are unchanged.

Checklist

Everyone

One topic.
Conventional Commit title: fix(build): honor selected external hot-function profiles.
Matching issue search completed; related open PRs reviewed, remaining issue search pending publication.
Measurements state the observed verdict and exact limitations.
Testing distinguishes tested, untested and required work.
Based on current main 9e6f231d, confirmed by the upstream compare page.
Complete tests/run_all.py result recorded; not run.
No game files, dumps, generated game files or build output.
Third-party sources untouched.
Existing Longshot source and authored fixtures; GPL-3.0-or-later.

When it applies

Source commit records Assisted-by: OpenAI Codex.
Existing switch behavior documented.
Observed run recorded under docs/tested/.
Behavioral regression fails on main and passes here.
Development documentation updated.

**Type:** fix · **Risk:** low · **Tested on:** Windows 11 Pro 26300, game-free · **Needs:** complete game-free suite and Linux verification ## Why An explicitly selected hot-function profile can silently lose to the build folder's default list. Editing an external profile at the same path can also reuse a translation produced from its old contents. ## How to reproduce **Seen on:** Windows 11 Pro 26300, Python 3.14.7; no GPU or game used. **Settings and switches:** external or explicitly empty `A64_HOT_FUNCS` with a different default `work/hot_funcs.txt`. **Steps:** Run the two selected override-consumer tests in `tests/test_make_game_cache.py` against main using its `LONGSHOT_CACHE_TEST_SOURCE` override, then against this branch. Run `check_selected_hot_list()` from `tests/test_lift_cache.py` against each implementation. **Expected:** the chosen list controls the real lifter consumer; empty disables it; editing selected contents invalidates the translation. **Actual:** main selects the default list for both overrides and does not invalidate on external-file edits. ## What changed Preserve explicit profile selection, including an empty value, and hash the selected profile's contents. Resolve relative profiles from the translation child's checkout root. - `nx2/make_game.py`: override precedence and selected-content lift-key inputs. - Tests exercise real environment merging, `a64_lift.hot_functions()`, cache invalidation, and cross-folder reuse. - Bind the cache fixture's real shipped-profile lookup to its authored checkout and close only its owned Windows lock. The profile-isolation part overlaps the fixture prerequisite in https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/592. - Scope is distinct from environment-key cleanup in https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/22 and profile-path object keys in https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/660. ## Measurements **Verdict:** the authored regressions fail on main and pass with the fix. No FPS claim. Hardware: Windows 11 Pro 26300, Intel Core i7-9700, 31.8 GiB usable RAM; pinned Python 3.14.7; no GPU work. Method: identical authored consumer/cache cases against main and the prepared branch. | Check | Before: main @ 9e6f231d | After: this PR @ b6d9ad2af4 | |---|---|---| | External override and explicitly empty override | Both fail: default function selected | Both pass | | Selected file edited at the same path | Cache key incorrectly retained | Lift-key regression passes | | Complete cache/consumer driver | Not claimed as a baseline full-driver result | 19 pass, 1 platform skip | Observed: `FAIL: editing the same external profile path changes the key` on main; main's two consumer assertions select `f_80000000` rather than the authored external function/empty set. Patched driver: `Ran 20 tests ... OK (skipped=1)`. Lift-key and cross-folder lift-store drivers pass. These checks ran on the source snapshot committed as b6d9ad2af4 before browser submission; no generated game files are included. ### Flags and settings `A64_HOT_FUNCS`: previously the builder's default list could override an explicit setting; now explicit selection wins, empty disables it, and unset retains existing behavior. No new switch/default or runtime quality change. ## Testing **Tested:** focused lift-key, complete cache/consumer, and lift-store drivers; authored baseline failure proofs. New run fragment is accepted by the actual tested-configurations renderer. **Not tested:** complete suite, native Linux, or an actual game build. The independent tested-configurations unit driver has a Windows CRLF-vs-LF pipe-fixture failure; it is not reported as passing. **Must be tested:** complete game-free suite on Windows and Linux, with source commits recorded, before readiness; by hand or the project's Test Farm. No runtime performance claim is proposed. ## Risks **What could break:** profile selection or cache invalidation could reuse the wrong translated code. **Turning it off:** revert the fix or leave the existing override unset. **Compatibility:** selected translation inputs can intentionally cause one retranslation. Saves, settings, shader/pipeline cache formats, renderer ABI and tools are unchanged. ## Checklist ### Everyone - [x] One topic. - [x] Conventional Commit title: `fix(build): honor selected external hot-function profiles`. - [ ] Matching issue search completed; related open PRs reviewed, remaining issue search pending publication. - [x] Measurements state the observed verdict and exact limitations. - [x] Testing distinguishes tested, untested and required work. - [x] Based on current main 9e6f231d, confirmed by the upstream compare page. - [ ] Complete `tests/run_all.py` result recorded; not run. - [x] No game files, dumps, generated game files or build output. - [x] Third-party sources untouched. - [x] Existing Longshot source and authored fixtures; GPL-3.0-or-later. ### When it applies - [x] Source commit records `Assisted-by: OpenAI Codex`. - [x] Existing switch behavior documented. - [x] Observed run recorded under docs/tested/. - [x] Behavioral regression fails on main and passes here. - [x] Development documentation updated.

siantm added 1 commit

2026-11-15 10:00:25 +00:00

fix(build): honor selected external hot-function profiles

CI / Conventional Commit title (pull_request) Blocked by required conditions

Details

CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Blocked by required conditions

Details

CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Blocked by required conditions

Details

CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Blocked by required conditions

Details

CI / Workflow lint (actionlint) (pull_request) Blocked by required conditions

Details

Guide screenshots / Guide images and local UI source review (pull_request) Blocked by required conditions

Details

Third-party parts / The same as upstream (pull_request) Blocked by required conditions

Details

Third-party parts / As their notes say (pull_request) Blocked by required conditions

Details

Website / Website on Cloudflare (pull_request) Blocked by required conditions

Details

PR description / Description follows the template (pull_request_target) Successful in 15s

Details

Website / Website on Cloudflare (pull_request_target) Has been skipped

Details

b6d9ad2af4

Extracted the focused local fixes from 27674af1 and 4e19ef16. Preserve explicit profile selection and selected-content cache invalidation; include authored consumer regressions and observed verification limits.

Assisted-by: OpenAI Codex