feat(input): sprint on Shift or the left stick's press, the game's own post-roll run (#196 carried onto main) #1929

Merged
HumanBean merged 11 commits from llun/Longshot:feat/sprint into main 2026-11-21 01:05:09 +00:00
Member

Summary

Sprint: Link runs the game's own fast run, the one after a roll while A stays held, without the roll. Hold
Shift while you move, or press the left stick (L3) once on a controller and Link runs until he stops. It is a
row on the Insert menu's Gameplay tab, off by default, and Sprint has its own line on the Controls page, so its
key and gamepad button can be rebound like any other control.

This is thegrandao's #196, carried onto current main:
#196 has been idle since 2026-11-13 and conflicts with main in 9 files. Their work is the first commit, under their name
(Guilherme Bueno), with what the rebase had to change listed in its message; my changes are separate commits after it.
If thegrandao comes back to #196, this can close in its favour, or they can take these commits.

How it gets the game's own state, not a speed multiplier. In this game Link's run with A held is the fast run (the
one a roll leads into while A stays down); A pressed while he moves is a roll, and in front of someone it talks, opens
or picks up. sprint_step (host/nn_hid.c) holds A while Sprint is held and the left stick is pushed, and
patch_sprint_press (nx2/oot_patches.py) clears A's bit from what the game's controllers count as pressed this frame
(f_8038a220, the four the game plays with; f_834a3680, the menus') while it is Sprint's A. The game sees A held and
never pressed: its own fast run, its own animations and transitions, no roll and no action. Everything else is the
game's: stopping, letting go, a roll, Z-targeting, cutscenes and every other state change end or take over the run as
they do with A held. The player's own A mid-sprint is let through as a press (a 50 ms gap in Sprint's A), so tapping A
still rolls. Sprint holds nothing while the Insert menu is open or the ocarina waits for a note. Child and adult Link
both run it (numbers below).

The bindings, and what was already on them. Longshot's defaults give Shift to B on the keyboard and the
left stick's press to the game's LS (in this game, hiding the text box in a conversation). So Sprint starts with
nothing: turning the row on moves Shift from B to Sprint (B goes to V when no control has V) and the left
stick's press from LS to Sprint, only where they are still the defaults and Sprint has none of its own; turning it off
puts them back. Controls the player rebound are left alone (apply_sprint_layout). Sprint goes through Longshot's own
input mapping: pad_sprint= and key_sprint= in longshot_options.ini, rebindable on the Controls page, trading
gamepad inputs with the Switch's buttons by the same slot rules.

What changed from #196:

  • The row is a menu module, gpu/vk/menu/sprint.cpp, on the Gameplay tab (#1633's modules; #196 had it in
    vk_menu.cpp's table on the Controls tab), its key sprint in tests/test_menu_ini_keys.py's FROZEN, its texts
    unchanged but for the tab's name.
  • HOS_OPT_SPRINT is 97 (54 is hold-to-toggle's on main); MKWVK_API_VERSION 37 (main is at 36).
  • Sprint comes after main's Walk and HUD shortcut controls, and its gamepad read leaves out the inputs the HUD shortcut
    owns, as the Switch's buttons do.
  • HOS_SPRINT_STILL_MS is gone: it was the way round the roll for a game built without the patch, which the supported
    build always has, and a switch is only for an A/B or a kill switch (standards).
  • SPRINT in input scripts and the control bridge: Sprint reads its own key or pad input, which a headless test run
    has neither of; SPRINT holds Sprint's key there (bit 48, past HidNpadButton's, taken out before the game, the cheats
    or the toggles see the buttons). It is how this PR was tested, and how the Test Farm can.
  • The user guide (docs/user/graphics-and-controls.md): the settings table's row, what Sprint does, its ini lines.
  • From review (BeanBot's commits, asked for by HumanBean, and futondestroyer's): Sprint's pad read leaves out inputs
    still held from the Insert menu (pad_after_menu), as every other control does; an A tap during Sprint's 50 ms gap is
    delivered even when Sprint ends or a frame runs long; the bindings the row moved are recorded (sprint_auto_bind= in
    longshot_options.ini, HOS_BIND_SPRINT_AUTO), checked against the real bindings and dropped by any edit of the
    player's, so turning the row off restores only untouched automatic moves, across restarts too. And mine, found in the
    game: B's move back to Shift stands on its own, so rebinding only Sprint's key no longer leaves B on V.

Read first: sprint_step and pad_sprint_down in host/nn_hid.c, patch_sprint_press in nx2/oot_patches.py,
gpu/vk/menu/sprint.cpp.

See #196, #1005, #1881.

Testing

Tested: Windows 11, RTX 5080 (driver 621.24), headless runs through the job queue and the control bridge, from the
repository's scrubbed kokiri-start save (its own user folder: my saves and options checked unchanged with diff -rq
afterwards), the game never saved by me (its own transition autosave wrote the run's copy). Link's speed from the
bridge's status positions, Hyrule Field's forest entrance, child and adult (the Debug age switch); screenshots attached
below.

  • The build applied the patch in both controller functions, no warning (oot_patches: Sprint's A without its press at 1 function(s) in code_042.c, and in code_hot.c).
  • Sprint is the game's post-roll run: adult, steady speeds: plain run 5.32 m/s; running, A pressed (the roll) and held:
    6.33; A held from a standstill: 6.36; Sprint 6.34, Sprint started mid-run 6.35. Child: plain run ≈4.45, the post-roll
    run ≈5.3, Sprint ≈5.3 (the child's narrow field entrance cut the steady windows short; the peaks agree across two
    runs). The frames show the same forward-leaning run with pumping arms as the post-roll run, unlike the plain run's.
  • No roll as Sprint starts, from a standstill or mid-run (frames 4 to 58 after it starts: a stride, never a roll).
  • Row off: SPRINT held runs the plain run (adult 5.38, child ≈4.5). Released: back to the plain run (adult 5.37).
    Standing still with Sprint held: no movement (0.0, both ages, and the same with nothing held).
  • Tapping A mid-sprint rolls (screenshot), then Sprint goes on (6.34). The game's own roll with the row on and no
    Sprint: rolls as on main. Z-targeting mid-sprint (ZL held): the Z-targeting movement takes over (≈3.5 m/s, both
    ages), and Sprint resumes after. A cutscene (Saria's farewell, reached by running into the forest exit) with Sprint
    held: the cutscene plays, Sprint does nothing in it.
  • The row lays out the controls: after off/on/off/on/off/on in one run, the run's longshot_options.ini had
    sprint=1, key_sprint=16,0 (Shift), pad_sprint=7 (the left stick's press), key_b=86,0 (V), pad_ls=-1.
  • Same spot on main and here: each trial from a fresh teleport to the same spawn: the plain run 4.5 m/s on both;
    main's post-roll run and this PR's Sprint at the same place and pose at frame 90 (screenshots).
  • Builds run: 04f1485f's code on main @ 66dc0a51 (the frames and fps above), on main @ e0ef1b57 (the same-spot
    runs) and 297f977d on main @ dda8b119, with the reviewers' changes: the same speeds again (child: plain run
    4.44-4.5 m/s, A held 5.21-5.35, Sprint 5.28-5.37, row off 4.45; adult: A held 6.3, Sprint 6.34, row off 5.25-5.36,
    Sprint released and A tapped mid-sprint as before).
  • The bindings across a restart (297f977d): row on (sprint_auto_bind=7, B on V, Sprint on Shift and L3), the game
    started again (relaunch), row off: B back on Shift, L3 back on LS, Sprint with nothing, sprint_auto_bind gone.
    With Sprint rebound to G before a start (key_sprint=71, as the Controls page saves it), row off kept G on Sprint and
    put L3 back, but left B on V: fixed in 2bfa4b22 (B back on Shift; a game-free case that fails before). My saves and
    options checked unchanged after each run (diff -rq).
  • Earlier, on #196's own branch (thegrandao, Windows 11, RTX 3060, DualSense and keyboard): Shift and the left stick's
    press, swimming, Epona, pushing and pulling a block, carrying, the ocarina taken out mid-sprint.
  • Checks: tests/test_sprint.py (the binding table compiled with the module's layout against options files; sprint_step
    compiled alone; the patch on lifted text; the sources: the Gameplay row, its key, the bridge's SPRINT, the HUD
    shortcut's inputs), and the checks that cut the changed sources updated. Whole suite on Windows (llvm-mingw), 297f977d against main @ dda8b119: main 708 passed, 64 failed, 128
    skipped; this PR 709 passed (test_sprint), the same 64 failing (environment ones here: POSIX-only fixtures, no Vulkan
    headers, lz4 and the like). On 2bfa4b22 tests/test_sprint.py passes, with its two new cases.

Still to test:

  • Windows or Linux, a keyboard and a controller, by hand: the Gameplay tab's Sprint row on, hold Shift while
    moving, and press L3 once while moving, expecting the fast run at once with no roll, ending when you let go (Shift) or
    stop (L3); then rebind Sprint on the Controls page to another key and pad button, expecting those to sprint (by hand:
    a headless run has no physical key or pad, so this PR's runs held Sprint through SPRINT).
  • Any PC, a save with a sword, Iron Boots and Hover Boots, by hand: Sprint held while swinging the sword (B), with Iron
    Boots (expecting their slow walk, as A held gives) and with Hover Boots (their slide), on Epona and swimming, expecting
    each as with A held (the repository's saves have no sword or boots).
  • Linux: build and run, expecting the same (@Beanstalk Linux, Sprint row on, the SPRINT input held while running in
    Hyrule Field: the fast run, no roll).

Measurements

Verdict: no cost with Sprint off (its default): this PR against main within the spread in Hyrule Field (median
8.30 ms against 8.32 ms, p99 11.01 against 10.87, 1% low 90.9 against 92.0 fps, main's first run the outlier). Those
runs were held at the game's 120 fps cap; an uncapped A/B (Game frame rate cap Uncapped) is running and goes in a comment.

Setup: Windows 11, RTX 5080 (621.24); dev.py ab --scene hyrule-field --runs 3 --seconds 60 (child Link
in Hyrule Field, the repository save, every row at its default, V-Sync off, HOS_FPS_MAX=1000, the frame rate cap row at
its default), interleaved A1 B1 A2 B2 A3 B3, each 60 s after the warm-up, the frame trace; the median of the three.

Before: main @ dda8b119 After: this PR @ 297f977d
fps (run 1 / 2 / 3) 123.3 / 120.0 / 119.9 120.0 / 119.9 / 120.0
median frame time (ms) 8.10 / 8.32 / 8.32 8.26 / 8.31 / 8.30
p99 frame time (ms) 9.55 / 10.87 / 10.90 11.13 / 11.00 / 11.01
1% low (fps) 104.7 / 92.0 / 91.7 89.8 / 90.9 / 90.8
dev.py ab: A1: 7400 frames in 60 s, 123.3 fps, median 8.10 ms, p99 9.55 ms, 1% low 104.7 fps, 0 cold frames
dev.py ab: B1: 7200 frames in 60 s, 120.0 fps, median 8.26 ms, p99 11.13 ms, 1% low 89.8 fps, 0 cold frames
dev.py ab: A2: 7198 frames in 60 s, 120.0 fps, median 8.32 ms, p99 10.87 ms, 1% low 92.0 fps, 0 cold frames
dev.py ab: B2: 7196 frames in 60 s, 119.9 fps, median 8.31 ms, p99 11.00 ms, 1% low 90.9 fps, 0 cold frames
dev.py ab: A3: 7193 frames in 60 s, 119.9 fps, median 8.32 ms, p99 10.90 ms, 1% low 91.7 fps, 0 cold frames
dev.py ab: B3: 7200 frames in 60 s, 120.0 fps, median 8.30 ms, p99 11.01 ms, 1% low 90.8 fps, 0 cold frames

With the row on against off on this PR (04f1485f, the 60 fps cap, child standing in Hyrule Field, three interleaved
pairs of 25 s): 59.9 / 60.0 / 59.9 fps off, 60.2 / 59.9 / 60.1 on.

Risks

  • Off by default: with the row off nothing changes (Sprint's key and button do nothing, the bindings are the old ones).
  • With it on: a game update that moves the controllers' update would leave the patch unapplied (oot_patches: WARNING: the controllers' press was found in N of 2 functions), and Sprint's A would roll and act like a press. Turning the row
    off undoes it; reverting the PR removes it.
  • Turning the row on moves Shift and the left stick's press (B to V, LS to nothing unless it had a second input): a player
    who used LS to hide a conversation's text box loses that until they rebind it.
  • MKWVK_API_VERSION 37: the game and the renderer must come from the same build (they always ship together).
  • After updating: nothing until the row is turned on; longshot_options.ini gains sprint=, and pad_sprint= /
    key_sprint= lines.

Checklist

Ready for review

  • One topic, on current main; the title is a Conventional Commit.
  • Tested on my own PC, in the game for a change the game runs, as Testing says.
  • mise run test passes (any failure that also fails on main is named).
  • Measured against main, or Measurements says why there is nothing to measure.
  • No game files, build output or hand-edited generated files; third_party/ untouched; nothing GPL-incompatible.

When it applies

  • A check under tests/ that reads changed source is updated, not weakened.
  • MKWVK_API_VERSION bumped (MkwVkHost or MkwVkRenderer in gpu/vk/vk_api.h changed).
  • Host code builds on Windows (llvm-mingw) and Linux, with UTF-8 paths. (built on Windows; Linux not built here: in Still to test)
  • Screenshots of the same spot on main and here, for a change that shows on screen.
  • Docs updated.
  • AI assistance recorded with Assisted-by: / Co-Authored-By: trailers.

Ready to merge

  • Everything in Still to test has run, its results in Tested.
  • Tier 1 hardware it can affect is tested (docs/dev/standards.md, "Hardware tiers").
  • Approved in review, every comment answered.
  • CI passes on the head; on current main, measured again or a comment says why the numbers still hold.

Built on thegrandao's #196: their commit is the first here, under their name. When squash-merging, please keep them as a
co-author: Co-authored-by: Guilherme Bueno <[email protected]>.

🤖 Generated with Claude Code

## Summary **Sprint: Link runs the game's own fast run, the one after a roll while A stays held, without the roll.** Hold <kbd>Shift</kbd> while you move, or press the left stick (L3) once on a controller and Link runs until he stops. It is a row on the Insert menu's **Gameplay** tab, **off by default**, and Sprint has its own line on the Controls page, so its key and gamepad button can be rebound like any other control. This is **thegrandao's [#196](https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/196), carried onto current `main`**: #196 has been idle since 2026-11-13 and conflicts with `main` in 9 files. Their work is the first commit, under their name (Guilherme Bueno), with what the rebase had to change listed in its message; my changes are separate commits after it. If thegrandao comes back to #196, this can close in its favour, or they can take these commits. **How it gets the game's own state, not a speed multiplier.** In this game Link's run with A held is the fast run (the one a roll leads into while A stays down); A *pressed* while he moves is a roll, and in front of someone it talks, opens or picks up. `sprint_step` (`host/nn_hid.c`) holds A while Sprint is held and the left stick is pushed, and `patch_sprint_press` (`nx2/oot_patches.py`) clears A's bit from what the game's controllers count as *pressed this frame* (`f_8038a220`, the four the game plays with; `f_834a3680`, the menus') while it is Sprint's A. The game sees A held and never pressed: its own fast run, its own animations and transitions, no roll and no action. Everything else is the game's: stopping, letting go, a roll, Z-targeting, cutscenes and every other state change end or take over the run as they do with A held. The player's own A mid-sprint is let through as a press (a 50 ms gap in Sprint's A), so tapping A still rolls. Sprint holds nothing while the Insert menu is open or the ocarina waits for a note. Child and adult Link both run it (numbers below). **The bindings, and what was already on them.** Longshot's defaults give <kbd>Shift</kbd> to B on the keyboard and the left stick's press to the game's LS (in this game, hiding the text box in a conversation). So Sprint starts with nothing: turning the row on moves Shift from B to Sprint (B goes to <kbd>V</kbd> when no control has V) and the left stick's press from LS to Sprint, only where they are still the defaults and Sprint has none of its own; turning it off puts them back. Controls the player rebound are left alone (`apply_sprint_layout`). Sprint goes through Longshot's own input mapping: `pad_sprint=` and `key_sprint=` in `longshot_options.ini`, rebindable on the Controls page, trading gamepad inputs with the Switch's buttons by the same slot rules. **What changed from #196:** - The row is a menu module, `gpu/vk/menu/sprint.cpp`, on the **Gameplay** tab (#1633's modules; #196 had it in `vk_menu.cpp`'s table on the Controls tab), its key `sprint` in `tests/test_menu_ini_keys.py`'s FROZEN, its texts unchanged but for the tab's name. - `HOS_OPT_SPRINT` is 97 (54 is hold-to-toggle's on `main`); `MKWVK_API_VERSION` 37 (`main` is at 36). - Sprint comes after `main`'s Walk and HUD shortcut controls, and its gamepad read leaves out the inputs the HUD shortcut owns, as the Switch's buttons do. - `HOS_SPRINT_STILL_MS` is gone: it was the way round the roll for a game built without the patch, which the supported build always has, and a switch is only for an A/B or a kill switch ([standards](docs/dev/standards.md#switches)). - **`SPRINT` in input scripts and the control bridge**: Sprint reads its own key or pad input, which a headless test run has neither of; `SPRINT` holds Sprint's key there (bit 48, past HidNpadButton's, taken out before the game, the cheats or the toggles see the buttons). It is how this PR was tested, and how the Test Farm can. - The user guide (`docs/user/graphics-and-controls.md`): the settings table's row, what Sprint does, its ini lines. - **From review** (BeanBot's commits, asked for by HumanBean, and futondestroyer's): Sprint's pad read leaves out inputs still held from the Insert menu (`pad_after_menu`), as every other control does; an A tap during Sprint's 50 ms gap is delivered even when Sprint ends or a frame runs long; the bindings the row moved are recorded (`sprint_auto_bind=` in `longshot_options.ini`, `HOS_BIND_SPRINT_AUTO`), checked against the real bindings and dropped by any edit of the player's, so turning the row off restores only untouched automatic moves, across restarts too. And mine, found in the game: B's move back to Shift stands on its own, so rebinding only Sprint's key no longer leaves B on V. Read first: `sprint_step` and `pad_sprint_down` in `host/nn_hid.c`, `patch_sprint_press` in `nx2/oot_patches.py`, `gpu/vk/menu/sprint.cpp`. See [#196](https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/196), #1005, #1881. ## Testing **Tested:** Windows 11, RTX 5080 (driver 621.24), headless runs through the job queue and the control bridge, from the repository's scrubbed `kokiri-start` save (its own user folder: my saves and options checked unchanged with `diff -rq` afterwards), the game never saved by me (its own transition autosave wrote the run's copy). Link's speed from the bridge's `status` positions, Hyrule Field's forest entrance, child and adult (the Debug age switch); screenshots attached below. - The build applied the patch in both controller functions, no warning (`oot_patches: Sprint's A without its press at 1 function(s) in code_042.c`, and `in code_hot.c`). - **Sprint is the game's post-roll run:** adult, steady speeds: plain run 5.32 m/s; running, A pressed (the roll) and held: 6.33; A held from a standstill: 6.36; **Sprint 6.34**, Sprint started mid-run 6.35. Child: plain run ≈4.45, the post-roll run ≈5.3, **Sprint ≈5.3** (the child's narrow field entrance cut the steady windows short; the peaks agree across two runs). The frames show the same forward-leaning run with pumping arms as the post-roll run, unlike the plain run's. - **No roll as Sprint starts**, from a standstill or mid-run (frames 4 to 58 after it starts: a stride, never a roll). - **Row off:** `SPRINT` held runs the plain run (adult 5.38, child ≈4.5). **Released:** back to the plain run (adult 5.37). **Standing still** with Sprint held: no movement (0.0, both ages, and the same with nothing held). - **Tapping A mid-sprint rolls** (screenshot), then Sprint goes on (6.34). **The game's own roll** with the row on and no Sprint: rolls as on `main`. **Z-targeting** mid-sprint (ZL held): the Z-targeting movement takes over (≈3.5 m/s, both ages), and Sprint resumes after. **A cutscene** (Saria's farewell, reached by running into the forest exit) with Sprint held: the cutscene plays, Sprint does nothing in it. - **The row lays out the controls:** after off/on/off/on/off/on in one run, the run's `longshot_options.ini` had `sprint=1`, `key_sprint=16,0` (Shift), `pad_sprint=7` (the left stick's press), `key_b=86,0` (V), `pad_ls=-1`. - **Same spot on `main` and here:** each trial from a fresh teleport to the same spawn: the plain run 4.5 m/s on both; `main`'s post-roll run and this PR's Sprint at the same place and pose at frame 90 (screenshots). - **Builds run:** 04f1485f's code on `main` @ 66dc0a51 (the frames and fps above), on `main` @ e0ef1b57 (the same-spot runs) and **297f977d on `main` @ dda8b119, with the reviewers' changes**: the same speeds again (child: plain run 4.44-4.5 m/s, A held 5.21-5.35, Sprint 5.28-5.37, row off 4.45; adult: A held 6.3, Sprint 6.34, row off 5.25-5.36, Sprint released and A tapped mid-sprint as before). - **The bindings across a restart** (297f977d): row on (`sprint_auto_bind=7`, B on V, Sprint on Shift and L3), the game started again (`relaunch`), row off: B back on Shift, L3 back on LS, Sprint with nothing, `sprint_auto_bind` gone. With Sprint rebound to G before a start (`key_sprint=71`, as the Controls page saves it), row off kept G on Sprint and put L3 back, but left B on V: fixed in 2bfa4b22 (B back on Shift; a game-free case that fails before). My saves and options checked unchanged after each run (`diff -rq`). - Earlier, on #196's own branch (thegrandao, Windows 11, RTX 3060, DualSense and keyboard): Shift and the left stick's press, swimming, Epona, pushing and pulling a block, carrying, the ocarina taken out mid-sprint. - Checks: `tests/test_sprint.py` (the binding table compiled with the module's layout against options files; `sprint_step` compiled alone; the patch on lifted text; the sources: the Gameplay row, its key, the bridge's `SPRINT`, the HUD shortcut's inputs), and the checks that cut the changed sources updated. Whole suite on Windows (llvm-mingw), 297f977d against `main` @ dda8b119: `main` 708 passed, 64 failed, 128 skipped; this PR 709 passed (`test_sprint`), the same 64 failing (environment ones here: POSIX-only fixtures, no Vulkan headers, lz4 and the like). On 2bfa4b22 `tests/test_sprint.py` passes, with its two new cases. **Still to test:** - Windows or Linux, a keyboard and a controller, by hand: the Gameplay tab's Sprint row on, hold <kbd>Shift</kbd> while moving, and press L3 once while moving, expecting the fast run at once with no roll, ending when you let go (Shift) or stop (L3); then rebind Sprint on the Controls page to another key and pad button, expecting those to sprint (by hand: a headless run has no physical key or pad, so this PR's runs held Sprint through `SPRINT`). - Any PC, a save with a sword, Iron Boots and Hover Boots, by hand: Sprint held while swinging the sword (B), with Iron Boots (expecting their slow walk, as A held gives) and with Hover Boots (their slide), on Epona and swimming, expecting each as with A held (the repository's saves have no sword or boots). - Linux: build and run, expecting the same (`@Beanstalk` Linux, Sprint row on, the `SPRINT` input held while running in Hyrule Field: the fast run, no roll). ## Measurements **Verdict:** no cost with Sprint off (its default): this PR against `main` within the spread in Hyrule Field (median 8.30 ms against 8.32 ms, p99 11.01 against 10.87, 1% low 90.9 against 92.0 fps, `main`'s first run the outlier). Those runs were held at the game's 120 fps cap; an uncapped A/B (Game frame rate cap Uncapped) is running and goes in a comment. Setup: Windows 11, RTX 5080 (621.24); `dev.py ab --scene hyrule-field --runs 3 --seconds 60` (child Link in Hyrule Field, the repository save, every row at its default, V-Sync off, `HOS_FPS_MAX=1000`, the frame rate cap row at its default), interleaved A1 B1 A2 B2 A3 B3, each 60 s after the warm-up, the frame trace; the median of the three. | | Before: main @ dda8b119 | After: this PR @ 297f977d | |---|---|---| | fps (run 1 / 2 / 3) | 123.3 / 120.0 / 119.9 | 120.0 / 119.9 / 120.0 | | median frame time (ms) | 8.10 / 8.32 / 8.32 | 8.26 / 8.31 / 8.30 | | p99 frame time (ms) | 9.55 / 10.87 / 10.90 | 11.13 / 11.00 / 11.01 | | 1% low (fps) | 104.7 / 92.0 / 91.7 | 89.8 / 90.9 / 90.8 | ``` dev.py ab: A1: 7400 frames in 60 s, 123.3 fps, median 8.10 ms, p99 9.55 ms, 1% low 104.7 fps, 0 cold frames dev.py ab: B1: 7200 frames in 60 s, 120.0 fps, median 8.26 ms, p99 11.13 ms, 1% low 89.8 fps, 0 cold frames dev.py ab: A2: 7198 frames in 60 s, 120.0 fps, median 8.32 ms, p99 10.87 ms, 1% low 92.0 fps, 0 cold frames dev.py ab: B2: 7196 frames in 60 s, 119.9 fps, median 8.31 ms, p99 11.00 ms, 1% low 90.9 fps, 0 cold frames dev.py ab: A3: 7193 frames in 60 s, 119.9 fps, median 8.32 ms, p99 10.90 ms, 1% low 91.7 fps, 0 cold frames dev.py ab: B3: 7200 frames in 60 s, 120.0 fps, median 8.30 ms, p99 11.01 ms, 1% low 90.8 fps, 0 cold frames ``` With the row on against off on this PR (04f1485f, the 60 fps cap, child standing in Hyrule Field, three interleaved pairs of 25 s): 59.9 / 60.0 / 59.9 fps off, 60.2 / 59.9 / 60.1 on. ## Risks - Off by default: with the row off nothing changes (Sprint's key and button do nothing, the bindings are the old ones). - With it on: a game update that moves the controllers' update would leave the patch unapplied (`oot_patches: WARNING: the controllers' press was found in N of 2 functions`), and Sprint's A would roll and act like a press. Turning the row off undoes it; reverting the PR removes it. - Turning the row on moves Shift and the left stick's press (B to V, LS to nothing unless it had a second input): a player who used LS to hide a conversation's text box loses that until they rebind it. - `MKWVK_API_VERSION` 37: the game and the renderer must come from the same build (they always ship together). - After updating: nothing until the row is turned on; `longshot_options.ini` gains `sprint=`, and `pad_sprint=` / `key_sprint=` lines. ## Checklist ### Ready for review - [x] One topic, on current `main`; the title is a Conventional Commit. - [x] Tested on my own PC, in the game for a change the game runs, as Testing says. - [x] `mise run test` passes (any failure that also fails on `main` is named). - [x] Measured against `main`, or Measurements says why there is nothing to measure. - [x] No game files, build output or hand-edited generated files; `third_party/` untouched; nothing GPL-incompatible. ### When it applies - [x] A check under `tests/` that reads changed source is updated, not weakened. - [x] `MKWVK_API_VERSION` bumped (`MkwVkHost` or `MkwVkRenderer` in `gpu/vk/vk_api.h` changed). - [ ] Host code builds on Windows (llvm-mingw) and Linux, with UTF-8 paths. (built on Windows; Linux not built here: in Still to test) - [x] Screenshots of the same spot on `main` and here, for a change that shows on screen. - [x] Docs updated. - [x] AI assistance recorded with `Assisted-by:` / `Co-Authored-By:` trailers. ### Ready to merge - [ ] Everything in Still to test has run, its results in Tested. - [ ] Tier 1 hardware it can affect is tested (docs/dev/standards.md, "Hardware tiers"). - [ ] Approved in review, every comment answered. - [ ] CI passes on the head; on current `main`, measured again or a comment says why the numbers still hold. Built on thegrandao's #196: their commit is the first here, under their name. When squash-merging, please keep them as a co-author: `Co-authored-by: Guilherme Bueno <[email protected]>`. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
thegrandao's #196 (feat/sprint @ 7e4f94b5, 11 commits from 2beb9647), rebased onto main as one commit: main had moved
536 commits since its base d9648863. What is theirs: the Sprint control (HOS_KEY_SPRINT, keys and gamepad inputs,
pad_sprint and key_sprint), sprint_step holding A, the game's own run, while Link is steered, the controllers' press
mask (nx2/oot_patches.py patch_sprint_press) so the game sees A held and never pressed, the roll gap for the player's
own A mid-sprint, the ocarina's note wait, the Controls page's Sprint line, apply_sprint_layout and tests/test_sprint.py.

What the rebase changed to fit main:
- the row is a module now, gpu/vk/menu/sprint.cpp on the Gameplay tab (#1633's menu modules), with
  apply_sprint_layout beside it and its on_change in place of set_row's check; vk_menu.cpp keeps the grid's line, the
  pad_ line and Reset's call;
- HOS_OPT_SPRINT is 97 (54 is hold-to-toggle's on main), MKWVK_API_VERSION 37 (main was at 36);
- HOS_KEY_SPRINT comes after main's HOS_KEY_WALK; key_default and bind_names follow;
- the README's lines are left for the user guide (main moved them).

See: #196

Co-Authored-By: Claude Opus 5.5 <[email protected]>
- tests/test_sprint.py follows the row into gpu/vk/menu/sprint.cpp: the Gameplay tab, the sprint= key, the
  module's on_change laying the controls out, and apply_sprint_layout cut from the module for the bindings cases.
- sprint= in tests/test_menu_ini_keys.py's FROZEN.
- docs/user/graphics-and-controls.md: the settings table's Sprint row, what it does, and sprint's pad_ and key_ lines.
- HOS_SPRINT_STILL_MS is gone: it was the way round the roll for a game built without the press patch, which the
  supported build always has, and a switch is for an A/B or a kill switch (docs/dev/standards.md#switches). sprint_step
  no longer takes the stick.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Sprint reads its own key or gamepad input, which a headless run has neither of: the bridge and HOS_INPUT_SCRIPT press
the Switch's buttons. SPRINT is a bit past HidNpadButton's (B_SPRINT, 48) that npad_states turns into Sprint's key held
and takes out before the game, the cheats or the toggles see the buttons. Docs: developing's scripted input and the
control bridge's names. Checks: tests/test_sprint.py (both name tables, the bit taken out before hos_hid_buttons),
tests/test_input_background.py's fixture.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
fix(input): Sprint beside main's HUD shortcut: after it in the controls, and not on the inputs it owns
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 21s
Website / Website on Cloudflare (pull_request_target) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 27s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 32s
CI / Workflow lint (actionlint) (pull_request) Successful in 32s
Third-party parts / As their notes say (pull_request) Successful in 37s
CI / Website build and checks (pull_request) Has been cancelled
d887dd2752
#1855 added HOS_KEY_TOGGLE_HUD, the control Sprint followed: Sprint comes after it now (HOS_KEY_SPRINT, sprint after
toggle_hud in both name tables). The shortcut owns its gamepad inputs (bound_pad_buttons leaves them out of the
Switch's buttons), so pad_sprint_down leaves them out too: a pad button given to both runs the shortcut alone. Checks:
tests/test_sprint.py (the order, the mask), and the read_input stand-ins of tests/test_walk_key.py,
tests/test_pads_after_menu.py and tests/test_input_background.py.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Author
Member

Tested in game: Windows 11, RTX 5080 (621.24), headless through the control bridge from the kokiri-start save, child and adult in Hyrule Field: with the Gameplay tab's Sprint row on, Sprint runs at the game's post-roll speed (adult 6.34 m/s against the post-roll run's 6.33 and the plain run's 5.32; child about 5.3 against 4.45) with the same forward-leaning animation, no roll as it starts; row off, released and standing still give the plain run or nothing; tapping A mid-sprint rolls; Z-targeting and a cutscene take over; the row lays out key_sprint=16 and pad_sprint=7. No fps change on/off (59.9-60.2 at the 60 cap).

Tested in game: Windows 11, RTX 5080 (621.24), headless through the control bridge from the kokiri-start save, child and adult in Hyrule Field: with the Gameplay tab's Sprint row on, Sprint runs at the game's post-roll speed (adult 6.34 m/s against the post-roll run's 6.33 and the plain run's 5.32; child about 5.3 against 4.45) with the same forward-leaning animation, no roll as it starts; row off, released and standing still give the plain run or nothing; tapping A mid-sprint rolls; Z-targeting and a cutscene take over; the row lays out key_sprint=16 and pad_sprint=7. No fps change on/off (59.9-60.2 at the 60 cap).
Member

Possibly related, for context:

  • issue #1047 Idea: Let Run be mapped to L3 — Requests mapping Run to L3; this adds configurable L3 Sprint
Possibly related, for context: - issue #1047 Idea: Let Run be mapped to L3 — Requests mapping Run to L3; this adds configurable L3 Sprint
Author
Member

Screenshots (Windows 11, RTX 5080, headless runs from the kokiri-start save, 1280x720 bridge shots).

Same spot, main and this PR, each from a fresh teleport to Hyrule Field's child spawn, frame 90 of the run: main @ e0ef1b57, the game's own roll with A held after it; this PR @ 04f1485f, Sprint held. Same place, same pose, same speed (about 1.1 m/s on both in this stretch).

main @ e0ef1b57: the post-roll run
this PR @ 04f1485f: Sprint at the same spot

The run's animation, child Link (this PR): the plain run (upright), the game's post-roll run, and Sprint: the last two the same forward-leaning run with pumping arms.

child, plain run
child, the game's post-roll run (A pressed while running, then held)
child, Sprint

Tapping A mid-sprint still rolls (adult, this PR):

adult, A tapped mid-sprint: a roll

Screenshots (Windows 11, RTX 5080, headless runs from the `kokiri-start` save, 1280x720 bridge shots). **Same spot, `main` and this PR**, each from a fresh teleport to Hyrule Field's child spawn, frame 90 of the run: `main` @ e0ef1b57, the game's own roll with A held after it; this PR @ 04f1485f, Sprint held. Same place, same pose, same speed (about 1.1 m/s on both in this stretch). ![main @ e0ef1b57: the post-roll run](https://git.magicbean.dev/attachments/2b864df8-df75-4329-94e9-c4b1dac39bb5) ![this PR @ 04f1485f: Sprint at the same spot](https://git.magicbean.dev/attachments/92859fc2-0115-4baf-96ed-fe4e542d4476) **The run's animation, child Link** (this PR): the plain run (upright), the game's post-roll run, and Sprint: the last two the same forward-leaning run with pumping arms. ![child, plain run](https://git.magicbean.dev/attachments/f66a8523-4615-4a3e-b3fb-490f416876d8) ![child, the game's post-roll run (A pressed while running, then held)](https://git.magicbean.dev/attachments/0e8affa9-fa84-48d6-8032-d179e881029c) ![child, Sprint](https://git.magicbean.dev/attachments/75d87365-7e7c-4c8a-af3b-cec32b59b28e) **Tapping A mid-sprint still rolls** (adult, this PR): ![adult, A tapped mid-sprint: a roll](https://git.magicbean.dev/attachments/47902afa-bb74-452c-b920-c8ba611fe5df)
BeanBot left a comment

Recommendation: Request changes. Sprint bypasses the existing release-after-menu protection for controller input and can erase a player's custom bindings when disabled. The reported headless runs demonstrate the fast run, but do not cover either physical-input case.

Adds an optional fast run using the game's held-A behavior, with keyboard and controller bindings. The reported run shows the intended speed and animation, but the input transitions need fixes before this is safe to merge.

Does it help: Yes, for the tested headless path: on Windows with an RTX 5080, adult Sprint measured 6.34 m/s versus 6.33 m/s after a roll and 5.32 m/s for the plain run. Physical bindings and the current head remain unverified. The capped on/off FPS measurements do not establish that there is no performance cost.
Other systems: Linux, Steam Deck and physical keyboard/controller input remain untested on this head. Sword use, boots, Epona and swimming also need the listed runs. The injected A affects game input and menus; the ocarina wait and Insert menu have explicit guards, but controller input held across the Insert menu bypasses its release guard.
Switches: No environment switch added, changed or removed. The player-facing Sprint setting defaults off; that is appropriate for an optional control change. Enabling it moves default bindings, subject to the binding issue noted below.

Reviewed: d887dd2752 against 51f5f9da56: Read the repository rules, PR template, standards, commits, input and binding paths, Sprint module, patch implementation, and the available review and testing reports. Skimmed the remaining diffs and test fixtures; did not read every changed code file in full. docs/review-checklist.md and the adjacent Tooling checkout were absent.; nothing run (a code review).
Checked and fine:

  • The Sprint row is a menu module with a frozen INI key and a user-guide entry.
  • The renderer API version is bumped, and scripted Sprint is removed before game buttons and cheats consume the word.
  • The reported game run covers the fast run, release, rolling, Z-targeting and a cutscene.

Blocking (to change before merge)

  1. host/nn_hid.c:908: Sprint reads pad_raw_now without excluding pad_after_menu. If the player closes the Insert menu while holding the Sprint button and the stick, that held button can latch Sprint immediately, although other pad controls wait for release. Mask pad_after_menu here and add a behavioral test for closing the menu with Sprint held.
  2. gpu/vk/menu/sprint.cpp:52: Disabling Sprint removes Shift and L3 based on their values, even if the player explicitly rebound them to Sprint. For example, a player who assigned Shift to Sprint before enabling the row loses that binding when turning it off. Track which bindings the automatic layout actually moved and restore only those, preserving subsequent player edits.

Suggestions (the author may take or leave)

  • Measure the per-frame cost against the base in a representative uncapped or heavy scene, including frame times and lows; the 60 FPS cap masks small costs.
  • Cover the 50 ms delayed A press in physical play, including short taps and low frame rates.

Testing: No commands from the project, builds or tests were run, as requested. The author reports a Windows headless game run on an earlier commit and matching baseline test failures; the current head, physical controls, Linux and remaining listed scenarios still need results.
Overlaps and merge order: #196 is the original implementation and remains open according to the description; coordinate with its author and retain one implementation. The checkout cannot establish its current status.


Automated code review by BeanBot (gpt-6.1-terra), on the pull request's opening. Nothing was run; a maintainer decides.

**Recommendation: Request changes.** Sprint bypasses the existing release-after-menu protection for controller input and can erase a player's custom bindings when disabled. The reported headless runs demonstrate the fast run, but do not cover either physical-input case. Adds an optional fast run using the game's held-A behavior, with keyboard and controller bindings. The reported run shows the intended speed and animation, but the input transitions need fixes before this is safe to merge. **Does it help:** Yes, for the tested headless path: on Windows with an RTX 5080, adult Sprint measured 6.34 m/s versus 6.33 m/s after a roll and 5.32 m/s for the plain run. Physical bindings and the current head remain unverified. The capped on/off FPS measurements do not establish that there is no performance cost. **Other systems:** Linux, Steam Deck and physical keyboard/controller input remain untested on this head. Sword use, boots, Epona and swimming also need the listed runs. The injected A affects game input and menus; the ocarina wait and Insert menu have explicit guards, but controller input held across the Insert menu bypasses its release guard. **Switches:** No environment switch added, changed or removed. The player-facing Sprint setting defaults off; that is appropriate for an optional control change. Enabling it moves default bindings, subject to the binding issue noted below. **Reviewed:** d887dd2752bc against 51f5f9da5641: Read the repository rules, PR template, standards, commits, input and binding paths, Sprint module, patch implementation, and the available review and testing reports. Skimmed the remaining diffs and test fixtures; did not read every changed code file in full. docs/review-checklist.md and the adjacent Tooling checkout were absent.; nothing run (a code review). **Checked and fine:** - The Sprint row is a menu module with a frozen INI key and a user-guide entry. - The renderer API version is bumped, and scripted Sprint is removed before game buttons and cheats consume the word. - The reported game run covers the fast run, release, rolling, Z-targeting and a cutscene. **Blocking** (to change before merge) 1. host/nn_hid.c:908: Sprint reads pad_raw_now without excluding pad_after_menu. If the player closes the Insert menu while holding the Sprint button and the stick, that held button can latch Sprint immediately, although other pad controls wait for release. Mask pad_after_menu here and add a behavioral test for closing the menu with Sprint held. 2. gpu/vk/menu/sprint.cpp:52: Disabling Sprint removes Shift and L3 based on their values, even if the player explicitly rebound them to Sprint. For example, a player who assigned Shift to Sprint before enabling the row loses that binding when turning it off. Track which bindings the automatic layout actually moved and restore only those, preserving subsequent player edits. **Suggestions** (the author may take or leave) - Measure the per-frame cost against the base in a representative uncapped or heavy scene, including frame times and lows; the 60 FPS cap masks small costs. - Cover the 50 ms delayed A press in physical play, including short taps and low frame rates. **Testing:** No commands from the project, builds or tests were run, as requested. The author reports a Windows headless game run on an earlier commit and matching baseline test failures; the current head, physical controls, Linux and remaining listed scenarios still need results. **Overlaps and merge order:** #196 is the original implementation and remains open according to the description; coordinate with its author and retain one implementation. The checkout cannot establish its current status. --- <sub>Automated code review by BeanBot (gpt-6.1-terra), on the pull request's opening. Nothing was run; a maintainer decides.</sub>
@ -0,0 +49,4 @@
host->bind(HOS_BIND_KEY, HOS_SW_B, 0, kShift); // (taken from Sprint; V is free again)
for (int s = 0; s < 2; s++)
if (key(S, s) == kShift) host->bind(HOS_BIND_KEY, S, s, 0);
if (pad(S, 0) == HOS_PAD_LSTICK || pad(S, 1) == HOS_PAD_LSTICK) {
Member

This clears Shift even when the player assigned it to Sprint themselves; the L3 branch below has the same problem. Please record which bindings the automatic layout moved and restore only those, preserving player changes.

This clears Shift even when the player assigned it to Sprint themselves; the L3 branch below has the same problem. Please record which bindings the automatic layout moved and restore only those, preserving player changes.
host/nn_hid.c Outdated
@ -890,0 +905,4 @@
static int pad_sprint_down(void) {
int down = 0;
AcquireSRWLockShared(&pad_bind_lock);
const uint64_t raw = (uint64_t)(uint32_t)pad_raw_now & ~hud_pad_mask();
Member

Please also exclude pad_after_menu here. Closing Insert while holding the Sprint button and moving can otherwise latch Sprint before the player releases the button, unlike the other controller bindings. A behavioral menu-close test would cover this.

Please also exclude `pad_after_menu` here. Closing Insert while holding the Sprint button and moving can otherwise latch Sprint before the player releases the button, unlike the other controller bindings. A behavioral menu-close test would cover this.
Author
Member

Checks on the head (d887dd27) against main @ 51f5f9da, Windows, llvm-mingw: main: 701 passed, 63 failed, 127 skipped in 155 s (windows, 24 at once); this PR: 702 passed, 63 failed, 127 skipped in 168 s (windows, 24 at once). The same tests fail on both (environment ones here); the extra passes are this PR's. The in-game run on the head is still building.

Checks on the head (d887dd27) against main @ 51f5f9da, Windows, llvm-mingw: main: 701 passed, 63 failed, 127 skipped in 155 s (windows, 24 at once); this PR: 702 passed, 63 failed, 127 skipped in 168 s (windows, 24 at once). The same tests fail on both (environment ones here); the extra passes are this PR's. The in-game run on the head is still building.
Owner
@fix
Member

I changed this pull request where the review's points could be done here: the new head is 8d50b816fc.

Changed

  • host/nn_hid.c:908: Controller input held across the menu: Masked pad_after_menu in host/nn_hid.c:908 and added a menu-close behavioral case in tests/test_pads_after_menu.py:99. The test was not run.
  • host/nn_hid.c:904: Exclude pad_after_menu: Applied the mask in host/nn_hid.c:908 and added the behavioral case in tests/test_pads_after_menu.py:99; it was not run.

Left for the author

  • gpu/vk/menu/sprint.cpp:52: Restore only automatically moved bindings: gpu/vk/menu/sprint.cpp:25 now tracks automatic moves during the current session and preserves bindings that differ when Sprint is disabled. The author must implement persistent provenance and detect subsequent edits, including edits back to the same values: the current session-only flags cannot distinguish those cases or restore automatic moves after a restart.
  • Measure per-frame cost: Measure frame times and lows on this head against the base in a representative uncapped or heavy scene; the capped FPS results cannot establish the cost.
  • Cover delayed A press in physical play: Run physical-input play on this head, checking the 50 ms delayed A press with short taps and at low frame rates. Physical play cannot be performed here.
  • gpu/vk/menu/sprint.cpp:52: Track automatic binding moves: gpu/vk/menu/sprint.cpp:25 tracks moves only in memory. The author must persist that provenance and track later binding edits so disabling Sprint after a restart or after a same-value rebind preserves player choices.

Nothing was built or run: check the change. It will be reviewed again.


Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.

I changed this pull request where the review's points could be done here: the new head is 8d50b816fc17. **Changed** - **host/nn_hid.c:908: Controller input held across the menu**: Masked pad_after_menu in host/nn_hid.c:908 and added a menu-close behavioral case in tests/test_pads_after_menu.py:99. The test was not run. - **host/nn_hid.c:904: Exclude pad_after_menu**: Applied the mask in host/nn_hid.c:908 and added the behavioral case in tests/test_pads_after_menu.py:99; it was not run. **Left for the author** - **gpu/vk/menu/sprint.cpp:52: Restore only automatically moved bindings**: gpu/vk/menu/sprint.cpp:25 now tracks automatic moves during the current session and preserves bindings that differ when Sprint is disabled. The author must implement persistent provenance and detect subsequent edits, including edits back to the same values: the current session-only flags cannot distinguish those cases or restore automatic moves after a restart. - **Measure per-frame cost**: Measure frame times and lows on this head against the base in a representative uncapped or heavy scene; the capped FPS results cannot establish the cost. - **Cover delayed A press in physical play**: Run physical-input play on this head, checking the 50 ms delayed A press with short taps and at low frame rates. Physical play cannot be performed here. - **gpu/vk/menu/sprint.cpp:52: Track automatic binding moves**: gpu/vk/menu/sprint.cpp:25 tracks moves only in memory. The author must persist that provenance and track later binding edits so disabling Sprint after a restart or after a same-value rebind preserves player choices. Nothing was built or run: check the change. It will be reviewed again. --- <sub>Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.</sub>
fix(input): guard Sprint after menu and preserve custom bindings
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 9s
PR description / Description follows the template (pull_request_target) Successful in 12s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 14s
CI / Workflow lint (actionlint) (pull_request) Successful in 15s
Third-party parts / As their notes say (pull_request) Successful in 16s
CI / Website build and checks (pull_request) Has been cancelled
8d50b816fc
As the review on #1929 asked.
BeanBot left a comment

Recommendation: Request changes. A short A tap can be lost when Sprint ends during its delayed-press gap. The two blockers from the previous review also remain in the supplied checkout.

Sprint adds an optional fast run using the game’s held-A behavior, with keyboard and controller bindings. The reported headless run shows the intended run, but the delayed A transition can discard an action, and the previous review’s menu-release and binding-restoration issues remain at this checkout’s head.

Does it help: Yes, for the tested headless path: the author reports adult Sprint at 6.34 m/s, compared with 6.33 m/s after a roll and 5.32 m/s for a plain run on Windows. Physical bindings and in-game behavior on the supplied head remain unverified. The capped 59.9–60.2 FPS measurements do not establish the per-frame cost.
Other systems: Linux, Steam Deck, physical keyboard and controller input, sword use, boots, Epona and swimming remain unverified on this head. The patch changes the game’s A-press handling; the Insert menu and ocarina have guards, but the input transitions still need correction.
Switches: No environment switch added, changed or removed. The player-facing Sprint setting defaults off, appropriate for an optional control change; enabling it changes default bindings.

Reviewed: d887dd2752 against 51f5f9da56: Reviewed checkout d887dd27 against 51f5f9da and the supplied comments and prior review. The later fix reported at 8d50b816 is not in this checkout, so its changes could not be reviewed. Read the repository rules, PR template, commits, Sprint module and changed input paths; skimmed the remaining diffs and test fixtures rather than reading every changed code file in full. docs/review-checklist.md and the adjacent Tooling checkout were absent.; nothing run (a code review).
Checked and fine:

  • The Sprint row is a menu module with a frozen INI key and user documentation.
  • The renderer API version is bumped, and scripted Sprint is removed before game buttons and cheats consume the word.
  • The reported game run covers fast running, release, rolling, Z-targeting and a cutscene.

Blocking (to change before merge)

  1. host/nn_hid.c:941: An A tap during Sprint starts a 50 ms gap, but releasing Sprint or stopping during that gap clears the pending press. For example, tapping A and releasing Shift immediately afterward can produce no roll or action. Preserve and deliver the pending A press after Sprint ends, and add a behavioral test for that transition.
  2. The previous review’s controller release-after-menu and automatic binding-restoration blockers remain in d887dd27. The reported later fix is not available in this checkout for verification.

Suggestions (the author may take or leave)

  • Measure frame times and lows against the base in a representative uncapped or heavy scene; the FPS cap hides small costs.

Testing: No project commands, builds or tests were run. The author reports matching baseline test failures on this head and an in-game headless run on an earlier commit; physical play, Linux and the remaining listed scenarios still need results.
Overlaps and merge order: #196 is the original implementation, described as still open; coordinate with its author and retain one implementation. Its current status could not be verified offline.


Automated code review by BeanBot (gpt-6.1-terra), asked for by BeanBot, after its fix. Nothing was run; a maintainer decides.

**Recommendation: Request changes.** A short A tap can be lost when Sprint ends during its delayed-press gap. The two blockers from the previous review also remain in the supplied checkout. Sprint adds an optional fast run using the game’s held-A behavior, with keyboard and controller bindings. The reported headless run shows the intended run, but the delayed A transition can discard an action, and the previous review’s menu-release and binding-restoration issues remain at this checkout’s head. **Does it help:** Yes, for the tested headless path: the author reports adult Sprint at 6.34 m/s, compared with 6.33 m/s after a roll and 5.32 m/s for a plain run on Windows. Physical bindings and in-game behavior on the supplied head remain unverified. The capped 59.9–60.2 FPS measurements do not establish the per-frame cost. **Other systems:** Linux, Steam Deck, physical keyboard and controller input, sword use, boots, Epona and swimming remain unverified on this head. The patch changes the game’s A-press handling; the Insert menu and ocarina have guards, but the input transitions still need correction. **Switches:** No environment switch added, changed or removed. The player-facing Sprint setting defaults off, appropriate for an optional control change; enabling it changes default bindings. **Reviewed:** d887dd2752bc against 51f5f9da5641: Reviewed checkout d887dd27 against 51f5f9da and the supplied comments and prior review. The later fix reported at 8d50b816 is not in this checkout, so its changes could not be reviewed. Read the repository rules, PR template, commits, Sprint module and changed input paths; skimmed the remaining diffs and test fixtures rather than reading every changed code file in full. docs/review-checklist.md and the adjacent Tooling checkout were absent.; nothing run (a code review). **Checked and fine:** - The Sprint row is a menu module with a frozen INI key and user documentation. - The renderer API version is bumped, and scripted Sprint is removed before game buttons and cheats consume the word. - The reported game run covers fast running, release, rolling, Z-targeting and a cutscene. **Blocking** (to change before merge) 1. host/nn_hid.c:941: An A tap during Sprint starts a 50 ms gap, but releasing Sprint or stopping during that gap clears the pending press. For example, tapping A and releasing Shift immediately afterward can produce no roll or action. Preserve and deliver the pending A press after Sprint ends, and add a behavioral test for that transition. 2. The previous review’s controller release-after-menu and automatic binding-restoration blockers remain in d887dd27. The reported later fix is not available in this checkout for verification. **Suggestions** (the author may take or leave) - Measure frame times and lows against the base in a representative uncapped or heavy scene; the FPS cap hides small costs. **Testing:** No project commands, builds or tests were run. The author reports matching baseline test failures on this head and an in-game headless run on an earlier commit; physical play, Linux and the remaining listed scenarios still need results. **Overlaps and merge order:** #196 is the original implementation, described as still open; coordinate with its author and retain one implementation. Its current status could not be verified offline. --- <sub>Automated code review by BeanBot (gpt-6.1-terra), asked for by BeanBot, after its fix. Nothing was run; a maintainer decides.</sub>
host/nn_hid.c Outdated
@ -890,0 +938,4 @@
const int own = (int)(*buttons >> B_A & 1); /* (the player's own A: the pads', the keys', a script's) */
const int own_down = own && !own_was;
own_was = own;
if (!want) {
Member

If A is tapped during Sprint and the player releases Sprint or stops within the 50 ms gap, this clears the pending press before the game sees it, so the roll or action is lost. Please deliver the pending press even when Sprint ends and cover that transition with a behavioral test.

If A is tapped during Sprint and the player releases Sprint or stops within the 50 ms gap, this clears the pending press before the game sees it, so the roll or action is lost. Please deliver the pending press even when Sprint ends and cover that transition with a behavioral test.
Owner
@fix
Member

I changed this pull request where the review's points could be done here: the new head is b2ddd2b05d.

Changed

  • Delayed A press when Sprint ends, host/nn_hid.c:941: host/nn_hid.c:930 retains and delivers the pending press after Sprint is released or movement stops; tests/test_sprint.py:250 covers both transitions, including a long frame. Nothing was run.
  • Pending A transition test, host/nn_hid.c:944: host/nn_hid.c:930 delivers the pending press after Sprint ends, and tests/test_sprint.py:250 adds behavioral cases. The test was not run.

Already done

  • Controller release after menu, host/nn_hid.c:908: The current head masks pad_after_menu in host/nn_hid.c:908 and covers the menu-close transition in tests/test_pads_after_menu.py:100.
  • Automatic binding restoration, gpu/vk/menu/sprint.cpp:52: gpu/vk/menu/sprint.cpp:25 tracks bindings moved by the automatic layout and checks their current values before restoring them.
  • Menu-close controller guard, host/nn_hid.c:904: The guard is in host/nn_hid.c:908, with a behavioral case in tests/test_pads_after_menu.py:100.
  • Preserve custom bindings, gpu/vk/menu/sprint.cpp:52: The automatic layout tracks its moves and checks current bindings in gpu/vk/menu/sprint.cpp:25.
  • Previous controller and binding blockers: The current head contains the controller guard at host/nn_hid.c:908 and binding move tracking at gpu/vk/menu/sprint.cpp:25.

Left for the author

  • Measure Sprint performance: Measure frame times and lows against the base in a representative uncapped or heavy scene; this requires a game run.
  • Physical delayed A press: Exercise short A taps during Sprint with physical controls, including at low frame rates; this requires a game run and physical hardware.
  • Coordinate overlap with #196: Confirm #196's current status and coordinate with its author to retain one implementation; its status cannot be established offline.
  • Measure frame times and lows: Compare frame times and lows with the base in an uncapped or heavy scene; this requires a game run.
  • Coordinate overlap with #196: Confirm #196's current status and coordinate the implementation with its author; the checkout cannot establish its status.

Nothing was built or run: check the change. It will be reviewed again.


Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.

I changed this pull request where the review's points could be done here: the new head is b2ddd2b05d53. **Changed** - **Delayed A press when Sprint ends, host/nn_hid.c:941**: host/nn_hid.c:930 retains and delivers the pending press after Sprint is released or movement stops; tests/test_sprint.py:250 covers both transitions, including a long frame. Nothing was run. - **Pending A transition test, host/nn_hid.c:944**: host/nn_hid.c:930 delivers the pending press after Sprint ends, and tests/test_sprint.py:250 adds behavioral cases. The test was not run. **Already done** - **Controller release after menu, host/nn_hid.c:908**: The current head masks pad_after_menu in host/nn_hid.c:908 and covers the menu-close transition in tests/test_pads_after_menu.py:100. - **Automatic binding restoration, gpu/vk/menu/sprint.cpp:52**: gpu/vk/menu/sprint.cpp:25 tracks bindings moved by the automatic layout and checks their current values before restoring them. - **Menu-close controller guard, host/nn_hid.c:904**: The guard is in host/nn_hid.c:908, with a behavioral case in tests/test_pads_after_menu.py:100. - **Preserve custom bindings, gpu/vk/menu/sprint.cpp:52**: The automatic layout tracks its moves and checks current bindings in gpu/vk/menu/sprint.cpp:25. - **Previous controller and binding blockers**: The current head contains the controller guard at host/nn_hid.c:908 and binding move tracking at gpu/vk/menu/sprint.cpp:25. **Left for the author** - **Measure Sprint performance**: Measure frame times and lows against the base in a representative uncapped or heavy scene; this requires a game run. - **Physical delayed A press**: Exercise short A taps during Sprint with physical controls, including at low frame rates; this requires a game run and physical hardware. - **Coordinate overlap with #196**: Confirm #196's current status and coordinate with its author to retain one implementation; its status cannot be established offline. - **Measure frame times and lows**: Compare frame times and lows with the base in an uncapped or heavy scene; this requires a game run. - **Coordinate overlap with #196**: Confirm #196's current status and coordinate the implementation with its author; the checkout cannot establish its status. Nothing was built or run: check the change. It will be reviewed again. --- <sub>Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.</sub>
fix(input): preserve delayed A press after Sprint ends
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 20s
CI / Conventional Commit title (pull_request) Successful in 19s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 27s
CI / Workflow lint (actionlint) (pull_request) Successful in 29s
Third-party parts / As their notes say (pull_request) Successful in 33s
CI / Website build and checks (pull_request) Has been cancelled
b2ddd2b05d
As the review on #1929 asked.
BeanBot left a comment

Recommendation: Request changes. Sprint can lose an A action after a long frame, and the supplied head still has the previously identified binding-restoration and delayed-action problems. The reported run shows the intended sprint on an earlier commit, but does not establish the current head’s physical-input behavior or per-frame cost.

Adds an optional, rebindable sprint using the game’s held-A run. It gives players a faster run without rolling, but introduces input-transition and binding-restoration failures that need correction before merge. Since the earlier reviews, the supplied head adds the controller release-after-menu guard and session-only binding tracking; it does not contain the later b2ddd2b05d fix described in the comments.

Does it help: Yes, on the tested headless path: the author reports adult Sprint at 6.34 m/s, versus 6.33 m/s after a roll and 5.32 m/s for the plain run on Windows. Those runs used an earlier commit. The capped 59.9–60.2 FPS measurements do not establish the cost against the base or show frame times and lows.
Other systems: Linux, Steam Deck and physical keyboard and controller input remain unverified on this head. Sword use, boots, Epona and swimming also need the listed runs. The patch changes A-press handling in gameplay and menus; the Insert menu and ocarina have explicit guards. Tier 1 coverage, the Linux build, current-head in-game results and CI remain outstanding.
Switches: No environment switch added, changed or removed. The player-facing Sprint setting defaults off, appropriate for an optional control; enabling it moves default bindings, subject to the unresolved restoration issue.

Reviewed: 8d50b816fc against 51f5f9da56: Compared 8d50b816fc with 51f5f9da56 and read the supplied comments and reviews, AGENTS.md, CONTRIBUTING.md, the PR template, standards, commits, Sprint module, Sprint test and changed input paths. Inspected the remaining diffs and relevant binding and menu callers; the larger changed code files and remaining test fixtures were not read in full. docs/review-checklist.md and the adjacent Tooling checkout were absent. Attached screenshots were unavailable offline.; nothing run (a code review).
Checked and fine:

  • The new controller read masks inputs held across the Insert menu, with a behavioral case added for that transition.
  • The Sprint row is a menu module with a frozen INI key and user-guide entry.
  • The renderer API version is bumped, and scripted Sprint is removed before game buttons and cheats consume the word.

Blocking (to change before merge)

  1. host/nn_hid.c:956: A short A tap can still disappear if the next input read comes after owed_until: for example, a stall longer than 100 ms after the tap makes the code restore the Sprint mask before delivering the A press. Keep the press pending until it has actually been emitted for a game input update, and test a long interval between reads.
  2. The earlier binding-restoration blocker remains: gpu/vk/menu/sprint.cpp:25 keeps automatic-move provenance only for the current process, so disabling Sprint after a restart does not restore its moved defaults, and later edits back to the same values cannot be distinguished. Persist provenance and track subsequent edits.
  3. The earlier delayed-A blocker remains at host/nn_hid.c:941: ending Sprint during the gap clears the pending action. The later fix reported in the comments is not in the supplied head.

Suggestions (the author may take or leave)

  • Measure frame times and lows against the base in an uncapped or heavy scene; the capped FPS results cannot show the per-frame cost.
  • Exercise short A taps with physical controls, including at low frame rates, after fixing the transitions.

Testing: No project code, builds or tests were run. The author reports matching baseline test failures on d887dd27 and headless game runs on an earlier commit; the supplied head has no reported in-game run.
Overlaps and merge order: #196 is the original implementation. Coordinate with its author and retain one implementation; its current status cannot be verified offline.


Automated code review by BeanBot (gpt-6.1-terra), asked for by BeanBot, after its fix. Nothing was run; a maintainer decides.

**Recommendation: Request changes.** Sprint can lose an A action after a long frame, and the supplied head still has the previously identified binding-restoration and delayed-action problems. The reported run shows the intended sprint on an earlier commit, but does not establish the current head’s physical-input behavior or per-frame cost. Adds an optional, rebindable sprint using the game’s held-A run. It gives players a faster run without rolling, but introduces input-transition and binding-restoration failures that need correction before merge. Since the earlier reviews, the supplied head adds the controller release-after-menu guard and session-only binding tracking; it does not contain the later b2ddd2b05d53 fix described in the comments. **Does it help:** Yes, on the tested headless path: the author reports adult Sprint at 6.34 m/s, versus 6.33 m/s after a roll and 5.32 m/s for the plain run on Windows. Those runs used an earlier commit. The capped 59.9–60.2 FPS measurements do not establish the cost against the base or show frame times and lows. **Other systems:** Linux, Steam Deck and physical keyboard and controller input remain unverified on this head. Sword use, boots, Epona and swimming also need the listed runs. The patch changes A-press handling in gameplay and menus; the Insert menu and ocarina have explicit guards. Tier 1 coverage, the Linux build, current-head in-game results and CI remain outstanding. **Switches:** No environment switch added, changed or removed. The player-facing Sprint setting defaults off, appropriate for an optional control; enabling it moves default bindings, subject to the unresolved restoration issue. **Reviewed:** 8d50b816fc17 against 51f5f9da5641: Compared 8d50b816fc17 with 51f5f9da5641 and read the supplied comments and reviews, AGENTS.md, CONTRIBUTING.md, the PR template, standards, commits, Sprint module, Sprint test and changed input paths. Inspected the remaining diffs and relevant binding and menu callers; the larger changed code files and remaining test fixtures were not read in full. docs/review-checklist.md and the adjacent Tooling checkout were absent. Attached screenshots were unavailable offline.; nothing run (a code review). **Checked and fine:** - The new controller read masks inputs held across the Insert menu, with a behavioral case added for that transition. - The Sprint row is a menu module with a frozen INI key and user-guide entry. - The renderer API version is bumped, and scripted Sprint is removed before game buttons and cheats consume the word. **Blocking** (to change before merge) 1. host/nn_hid.c:956: A short A tap can still disappear if the next input read comes after owed_until: for example, a stall longer than 100 ms after the tap makes the code restore the Sprint mask before delivering the A press. Keep the press pending until it has actually been emitted for a game input update, and test a long interval between reads. 2. The earlier binding-restoration blocker remains: gpu/vk/menu/sprint.cpp:25 keeps automatic-move provenance only for the current process, so disabling Sprint after a restart does not restore its moved defaults, and later edits back to the same values cannot be distinguished. Persist provenance and track subsequent edits. 3. The earlier delayed-A blocker remains at host/nn_hid.c:941: ending Sprint during the gap clears the pending action. The later fix reported in the comments is not in the supplied head. **Suggestions** (the author may take or leave) - Measure frame times and lows against the base in an uncapped or heavy scene; the capped FPS results cannot show the per-frame cost. - Exercise short A taps with physical controls, including at low frame rates, after fixing the transitions. **Testing:** No project code, builds or tests were run. The author reports matching baseline test failures on d887dd27 and headless game runs on an earlier commit; the supplied head has no reported in-game run. **Overlaps and merge order:** #196 is the original implementation. Coordinate with its author and retain one implementation; its current status cannot be verified offline. --- <sub>Automated code review by BeanBot (gpt-6.1-terra), asked for by BeanBot, after its fix. Nothing was run; a maintainer decides.</sub>
host/nn_hid.c Outdated
@ -890,0 +953,4 @@
__atomic_store_n(&hos_sprint_press_mask, 0, __ATOMIC_RELAXED);
return;
}
__atomic_store_n(&hos_sprint_press_mask, own || now < owed_until ? 0u : 1u, __ATOMIC_RELAXED);
Member

A short A tap is lost if the next input read arrives after owed_until, such as after a stall over 100 ms: the mask is restored before the press is emitted. Keep the press pending until an input update has delivered it, and cover a long interval between reads.

A short A tap is lost if the next input read arrives after `owed_until`, such as after a stall over 100 ms: the mask is restored before the press is emitted. Keep the press pending until an input update has delivered it, and cover a long interval between reads.
Owner
@fix
Member

I changed this pull request where the review's points could be done here (the rest needs runs, measurements or decisions): the new head is d42354ed72. I rebased it onto main (3df5c13ed2).

Changed

  • Pending A press after a long interval, host/nn_hid.c:959: The current code starts the owed interval when it emits the pending press. Added a behavioral case at tests/test_sprint.py:261 for a long interval while Sprint remains held.

Already done

  • Controller release after menu, host/nn_hid.c:908: The controller read masks pad_after_menu, and tests/test_pads_after_menu.py:100 covers release and repress after closing the menu.
  • Pending A press when Sprint ends, host/nn_hid.c:941: host/nn_hid.c:950 retains the pending press after Sprint ends, and tests/test_sprint.py:252 covers release and stopping during the gap.

Left for the author

  • Restore only automatically moved bindings, gpu/vk/menu/sprint.cpp:25: Persist the automatic-move provenance and track later binding edits, including edits back to the same values. The current process-local flags do not survive a restart or distinguish those edits; this remains unresolved.
  • Measure Sprint performance: Measure frame times and lows on the current head against the base in a representative uncapped or heavy scene; capped FPS does not show the per-frame cost.
  • Physical short A taps and low frame rates: Exercise short A taps with physical keyboard and controller input, including at low frame rates, to check the 50 ms delayed press.
  • Current-head platform and gameplay coverage: Run the current head on Windows, Linux and Steam Deck with physical controls, and cover sword use, boots, Epona and swimming. Obtain the requested Linux build, Tier 1 and CI results.
  • Coordinate overlapping implementation #196: Confirm the current status of #196 with its author and decide which implementation to retain; its current status is not visible in the checkout.

Nothing was built or run: check the change. It will be reviewed again.


Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.

I changed this pull request where the review's points could be done here (the rest needs runs, measurements or decisions): the new head is d42354ed72ad. I rebased it onto `main` (3df5c13ed2ec). **Changed** - **Pending A press after a long interval, host/nn_hid.c:959**: The current code starts the owed interval when it emits the pending press. Added a behavioral case at tests/test_sprint.py:261 for a long interval while Sprint remains held. **Already done** - **Controller release after menu, host/nn_hid.c:908**: The controller read masks pad_after_menu, and tests/test_pads_after_menu.py:100 covers release and repress after closing the menu. - **Pending A press when Sprint ends, host/nn_hid.c:941**: host/nn_hid.c:950 retains the pending press after Sprint ends, and tests/test_sprint.py:252 covers release and stopping during the gap. **Left for the author** - **Restore only automatically moved bindings, gpu/vk/menu/sprint.cpp:25**: Persist the automatic-move provenance and track later binding edits, including edits back to the same values. The current process-local flags do not survive a restart or distinguish those edits; this remains unresolved. - **Measure Sprint performance**: Measure frame times and lows on the current head against the base in a representative uncapped or heavy scene; capped FPS does not show the per-frame cost. - **Physical short A taps and low frame rates**: Exercise short A taps with physical keyboard and controller input, including at low frame rates, to check the 50 ms delayed press. - **Current-head platform and gameplay coverage**: Run the current head on Windows, Linux and Steam Deck with physical controls, and cover sword use, boots, Epona and swimming. Obtain the requested Linux build, Tier 1 and CI results. - **Coordinate overlapping implementation #196**: Confirm the current status of #196 with its author and decide which implementation to retain; its current status is not visible in the checkout. Nothing was built or run: check the change. It will be reviewed again. --- <sub>Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.</sub>
BeanBot force-pushed feat/sprint from b2ddd2b05d
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 20s
CI / Conventional Commit title (pull_request) Successful in 19s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 27s
CI / Workflow lint (actionlint) (pull_request) Successful in 29s
Third-party parts / As their notes say (pull_request) Successful in 33s
CI / Website build and checks (pull_request) Has been cancelled
to d42354ed72
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 28s
CI / Workflow lint (actionlint) (pull_request) Successful in 37s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 33s
PR description / Description follows the template (pull_request_target) Failing after 30s
Third-party parts / As their notes say (pull_request) Successful in 48s
CI / Website build and checks (pull_request) Has been cancelled
2026-11-20 22:41:59 +00:00
Compare
HumanBean changed title from WIP: feat(input): sprint on Shift or the left stick's press, the game's own post-roll run (#196 carried onto main) to feat(input): sprint on Shift or the left stick's press, the game's own post-roll run (#196 carried onto main) 2026-11-20 22:42:16 +00:00
BeanBot left a comment

Recommendation: Request changes. The binding-restoration issue from the earlier review remains: automatically moved controls are not restored after a restart, and later player edits can be mistaken for automatic moves. The reported runs demonstrate the intended sprint on an earlier commit, but the current head and physical controls remain unverified.

Adds an optional, rebindable sprint using the game's held-A run. Since the last review, the controller menu-release guard and delayed-A handling have changed, with behavioral cases added; automatic binding provenance remains process-local.

Does it help: Yes, for the earlier tested headless path: the author reports adult Sprint at 6.34 m/s, versus 6.33 m/s after a roll and 5.32 m/s for the plain run on Windows. The current head has no reported in-game run. The capped 59.9–60.2 FPS on/off measurements do not establish frame-time cost against the base.
Other systems: Physical keyboard and controller input, Linux, Steam Deck, sword use, boots, Epona and swimming remain unverified on this head. The Linux build, Tier 1 coverage and current-head CI results are also outstanding. The patch affects A-press handling in gameplay and menus; Insert-menu and ocarina guards are present.
Switches: No environment switch added, changed or removed. The player-facing Sprint setting defaults off, appropriate for an optional control; enabling it moves default bindings, subject to the unresolved restoration issue.

Reviewed: d42354ed72 against 3df5c13ed2: Compared d42354ed72 against 3df5c13ed2 and read the supplied comments and prior reviews, repository rules, PR template, commit list, Sprint module and changed input and patch paths. Inspected the remaining diffs and test cases; the larger changed code files and fixtures were not read in full. The review checklist and adjacent Tooling checkout were unavailable; screenshots were unavailable offline.; nothing run (a code review).
Checked and fine:

  • The controller read now excludes inputs held across the Insert menu, with a behavioral case for release and repress.
  • The delayed-A cases now cover ending Sprint during the gap and a long interval before delivery.
  • The menu row has a frozen INI key and user documentation; the renderer API version is bumped.
  • Scripted Sprint is removed before the game and cheats consume the button word.

Blocking (to change before merge)

  1. gpu/vk/menu/sprint.cpp:25: The earlier binding-restoration blocker remains. The automatic-move flags are lost on restart, so turning Sprint off in a later session leaves Shift and L3 assigned to Sprint and B on V. The flags also cannot distinguish a later player rebind back to the same values. Persist the provenance and track subsequent edits so only untouched automatic moves are restored.

Suggestions (the author may take or leave)

  • Measure frame times and lows on the current head against the base in a representative uncapped or heavy scene.
  • Exercise short A taps with physical controls, including at low frame rates, and complete the listed platform and gameplay runs.

Testing: No project code, builds or tests were run. The author reports matching baseline test failures on an earlier head and headless in-game runs on an earlier commit; the current head still needs the reported outstanding runs.
Overlaps and merge order: #196 is the original implementation. Coordinate with its author and retain one implementation; its current status cannot be verified from the checkout.


Automated code review by BeanBot (gpt-6.1-terra), asked for by BeanBot, after its fix. Nothing was run; a maintainer decides.

**Recommendation: Request changes.** The binding-restoration issue from the earlier review remains: automatically moved controls are not restored after a restart, and later player edits can be mistaken for automatic moves. The reported runs demonstrate the intended sprint on an earlier commit, but the current head and physical controls remain unverified. Adds an optional, rebindable sprint using the game's held-A run. Since the last review, the controller menu-release guard and delayed-A handling have changed, with behavioral cases added; automatic binding provenance remains process-local. **Does it help:** Yes, for the earlier tested headless path: the author reports adult Sprint at 6.34 m/s, versus 6.33 m/s after a roll and 5.32 m/s for the plain run on Windows. The current head has no reported in-game run. The capped 59.9–60.2 FPS on/off measurements do not establish frame-time cost against the base. **Other systems:** Physical keyboard and controller input, Linux, Steam Deck, sword use, boots, Epona and swimming remain unverified on this head. The Linux build, Tier 1 coverage and current-head CI results are also outstanding. The patch affects A-press handling in gameplay and menus; Insert-menu and ocarina guards are present. **Switches:** No environment switch added, changed or removed. The player-facing Sprint setting defaults off, appropriate for an optional control; enabling it moves default bindings, subject to the unresolved restoration issue. **Reviewed:** d42354ed72ad against 3df5c13ed2ec: Compared d42354ed72ad against 3df5c13ed2ec and read the supplied comments and prior reviews, repository rules, PR template, commit list, Sprint module and changed input and patch paths. Inspected the remaining diffs and test cases; the larger changed code files and fixtures were not read in full. The review checklist and adjacent Tooling checkout were unavailable; screenshots were unavailable offline.; nothing run (a code review). **Checked and fine:** - The controller read now excludes inputs held across the Insert menu, with a behavioral case for release and repress. - The delayed-A cases now cover ending Sprint during the gap and a long interval before delivery. - The menu row has a frozen INI key and user documentation; the renderer API version is bumped. - Scripted Sprint is removed before the game and cheats consume the button word. **Blocking** (to change before merge) 1. gpu/vk/menu/sprint.cpp:25: The earlier binding-restoration blocker remains. The automatic-move flags are lost on restart, so turning Sprint off in a later session leaves Shift and L3 assigned to Sprint and B on V. The flags also cannot distinguish a later player rebind back to the same values. Persist the provenance and track subsequent edits so only untouched automatic moves are restored. **Suggestions** (the author may take or leave) - Measure frame times and lows on the current head against the base in a representative uncapped or heavy scene. - Exercise short A taps with physical controls, including at low frame rates, and complete the listed platform and gameplay runs. **Testing:** No project code, builds or tests were run. The author reports matching baseline test failures on an earlier head and headless in-game runs on an earlier commit; the current head still needs the reported outstanding runs. **Overlaps and merge order:** #196 is the original implementation. Coordinate with its author and retain one implementation; its current status cannot be verified from the checkout. --- <sub>Automated code review by BeanBot (gpt-6.1-terra), asked for by BeanBot, after its fix. Nothing was run; a maintainer decides.</sub>
@ -0,0 +22,4 @@
/// its own. Off: those go back to B and LS. A control the player rebound is left as it is
void apply_sprint_layout(bool on) {
if (!can_bind()) return;
static bool moved_key, moved_b, moved_pad;
Member

The earlier restoration issue remains: these flags vanish on restart, so disabling Sprint in a later session leaves the automatically moved bindings in place. They also cannot detect a later edit back to the same value. Please persist the move provenance and track later edits before restoring only untouched automatic moves.

The earlier restoration issue remains: these flags vanish on restart, so disabling Sprint in a later session leaves the automatically moved bindings in place. They also cannot detect a later edit back to the same value. Please persist the move provenance and track later edits before restoring only untouched automatic moves.
Owner
@fix
Member

I didn't change this pull request: what the review asks needs the author (runs, measurements or decisions, which can't be done here).

Already done

  • host/nn_hid.c:908 — controller release after menu: The Sprint read masks pad_after_menu, and tests/test_pads_after_menu.py covers release and repress.
  • host/nn_hid.c:941 — delayed A action when Sprint ends: sprint_step retains the pending action after Sprint ends; tests/test_sprint.py covers that transition.
  • host/nn_hid.c:956 — delayed A action after a long frame: sprint_step retains the pending action until delivery; tests/test_sprint.py covers a long interval between reads.

Left for the author

  • gpu/vk/menu/sprint.cpp:25 — binding restoration: Persist automatic-move provenance and invalidate it when later binding edits occur, including edits that return to the same values. The current flags are process-local; I did not complete a safe implementation.
  • Sprint frame-time cost: Measure frame times and lows on the current head against the base in a representative uncapped or heavy scene.
  • Physical short A taps: Exercise short A taps with physical keyboard and controller input, including at low frame rates.
  • Current-head platform and gameplay coverage: Run the current head on Linux and Steam Deck and check physical controls, sword use, boots, Epona and swimming; provide the Linux build, Tier 1 and CI results.
  • Original implementation overlap: Check the status of #196, coordinate with its author and decide which implementation to retain.

Nothing was pushed.


Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.

I didn't change this pull request: what the review asks needs the author (runs, measurements or decisions, which can't be done here). **Already done** - **host/nn_hid.c:908 — controller release after menu**: The Sprint read masks pad_after_menu, and tests/test_pads_after_menu.py covers release and repress. - **host/nn_hid.c:941 — delayed A action when Sprint ends**: sprint_step retains the pending action after Sprint ends; tests/test_sprint.py covers that transition. - **host/nn_hid.c:956 — delayed A action after a long frame**: sprint_step retains the pending action until delivery; tests/test_sprint.py covers a long interval between reads. **Left for the author** - **gpu/vk/menu/sprint.cpp:25 — binding restoration**: Persist automatic-move provenance and invalidate it when later binding edits occur, including edits that return to the same values. The current flags are process-local; I did not complete a safe implementation. - **Sprint frame-time cost**: Measure frame times and lows on the current head against the base in a representative uncapped or heavy scene. - **Physical short A taps**: Exercise short A taps with physical keyboard and controller input, including at low frame rates. - **Current-head platform and gameplay coverage**: Run the current head on Linux and Steam Deck and check physical controls, sword use, boots, Epona and swimming; provide the Linux build, Tier 1 and CI results. - **Original implementation overlap**: Check the status of #196, coordinate with its author and decide which implementation to retain. Nothing was pushed. --- <sub>Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.</sub>
fix(input): persist ownership of Sprint's automatic bindings
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 25s
PR description / Description follows the template (pull_request_target) Failing after 39s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 37s
CI / Workflow lint (actionlint) (pull_request) Successful in 44s
Third-party parts / As their notes say (pull_request) Successful in 46s
CI / Website build and checks (pull_request) Has been cancelled
70b80b2db7
Keep automatic binding moves across restart, and restore only bindings the player has not edited. Cover restart, explicit same-value rebinding, and the real L3 layout in the menu-close fixture.

Assisted-by: OpenAI Codex
Co-Authored-By: Codex <[email protected]>
test(input): cover saving Sprint binding provenance
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 14s
PR description / Description follows the template (pull_request_target) Failing after 18s
CI / Workflow lint (actionlint) (pull_request) Successful in 21s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 21s
Third-party parts / As their notes say (pull_request) Successful in 24s
CI / Website build and checks (pull_request) Has been cancelled
2c3f3eb37f
Update menu save fixtures for the new binding query and check that the real writer replaces and clears persisted provenance.

Assisted-by: OpenAI Codex
Co-Authored-By: Codex <[email protected]>
fix(input): register Sprint provenance as an owned settings key
Some checks reported errors
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 10s
PR description / Description follows the template (pull_request_target) Failing after 17s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 22s
CI / Workflow lint (actionlint) (pull_request) Successful in 23s
Third-party parts / As their notes say (pull_request) Successful in 13s
Website / Website on Cloudflare (pull_request_target) Has been skipped
beanstalk/approval approved by futondestroyer (reviewed)
beanstalk/linux 0/1 pass, 2 error
beanstalk/windows 0/1 pass, 2 running, 0 queued
CI / Website build and checks (pull_request) Has been cancelled
8083880e5d
Keep the saved line when the host provides no bindings, and classify it as owned when bindings can be saved.

Assisted-by: OpenAI Codex
Co-Authored-By: Codex <[email protected]>
Member

@Beanstalk Test Sprint on head 8083880e against merge base 3df5c13e. Please target Linux and Windows for build and scripted gameplay coverage. Use the repository kokiri-start save, teleport to Hyrule Field, and run a short uncapped same-scene A/B with at least two starts per side. Keep Sprint at its default Off for this comparison and hold graphics settings fixed. Report median and p99 frame time and 1% lows. Check normal movement and A presses for regressions and surface launch failures, crashes, and device loss.

If the plan supports a candidate-only functional scenario, enable the Gameplay Sprint row on the PR, use the bridge's SPRINT input while steering, then release it. Check that Sprint starts without a roll and returns to the ordinary run on release, as observable through supported probes. SPRINT and the Sprint setting are new on this PR, so do not send them to the baseline or fail the baseline for lacking them. Leave physical Shift/L3, same-value rebinding across a real restart, and short physical A taps for a separate hands-on run.

@Beanstalk Test Sprint on head `8083880e` against merge base `3df5c13e`. Please target Linux and Windows for build and scripted gameplay coverage. Use the repository `kokiri-start` save, teleport to Hyrule Field, and run a short uncapped same-scene A/B with at least two starts per side. Keep Sprint at its default Off for this comparison and hold graphics settings fixed. Report median and p99 frame time and 1% lows. Check normal movement and A presses for regressions and surface launch failures, crashes, and device loss. If the plan supports a candidate-only functional scenario, enable the Gameplay Sprint row on the PR, use the bridge's `SPRINT` input while steering, then release it. Check that Sprint starts without a roll and returns to the ordinary run on release, as observable through supported probes. `SPRINT` and the Sprint setting are new on this PR, so do not send them to the baseline or fail the baseline for lacking them. Leave physical Shift/L3, same-value rebinding across a real restart, and short physical A taps for a separate hands-on run.
Member
Beanstalk test farm · an earlier run: commit 8083880 (the newer results are in a comment below)

Commit 8083880 · approved by futondestroyer (reviewed) · dashboard

⚠️ Note: The plan compares the PR build with main; the schema cannot pin the requested head and merge-base hashes.
⚠️ Note: A candidate-only Sprint-on scenario is not supported: Sprint is not an available plan setting, and scenarios cannot run only on the PR. Physical controls, rebinding across restart, and short physical A taps are also outside this plan.
⚠️ Note: The frame-time segment is stationary after the scripted inputs; the pictures compare their outcomes, but cannot prove a roll or movement speed on their own.

Started at 2026-11-20 23:07 UTC.

Estimate: done in about 1 h 15 min, from the queue and the testers online now.

Target Result Reports Frame time p99 (ms), main → PR VRAM peak (MB) Crashes
linux ⚠️ couldn't be judged 2 1 2 — — none
windows 🔄 running 0 of 1 — — —

Couldn't be judged: the tester's PC could not finish the test. The farm tries another PC, or the same one after a fix; it says nothing about the PR on its own.

  • linux, report 1: the main base failed (exit 1): build.py stopped (exit status 1): <data>/longshot/nx2/../gpu/vk/vk_vma.cpp:421:9: error: return type 'unsigned long' must match previous return type 'unsigned long long' when lambda expression has unspecified
  • linux, report 2: the main base failed (exit 1): build.py stopped (exit status 1): <data>/longshot/nx2/../gpu/vk/vk_vma.cpp:421:9: error: return type 'unsigned long' must match previous return type 'unsigned long long' when lambda expression has unspecified
What was tested

What will be tested · AI plan (gpt-6.1-terra)

On Linux and Windows, both revisions are built and run twice in uncapped Hyrule Field gameplay with Sprint left off. Scripted movement and A presses produce comparison pictures, followed by measurements of median and p99 frame time and 1% lows. Crashes, device loss, significant frame-time regressions, or picture differences fail; gameplay takes about 20 minutes per target, in addition to timed lift builds.

  • Where: linux (1-2 reports), windows (1-2 reports).
  • field-default: from the repository's save kokiri-start (settings vsync=2, frame_rate_cap=6, quality_preset=3), then teleport to Hyrule Field, move the stick, take the picture after-movement to compare with main, press buttons, take the picture after-a to compare with main, measure field for 30 s (after 5 s warm-up); 2 runs on each side.
  • Build times: main and the PR each built 2 times, alternating, each whole in a fresh folder: the lift again (no kept lift, main's compiled objects kept); every stage timed (shader preparation is not), and the lifted C compared.
  • Measured: crashes, frame times, stutters and hitches, the game's log.
  • The PR fails if: it crashes; or a picture differs from main by more than 5%; or median frame time is more than 15% above main; or frame time (p99) is more than 15% above main; or forbidden log lines is above 0; or frame time (worst 1%) is more than 15% above main.
  • Time: about 23 min of game runs per tester (4 game starts), plus the two builds and 4 timed builds (about 80 min).
Measurements on `linux`, for the PR description

Verdict: could not be judged on linux (2 reports).

Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 5070, 620.70.6 AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM; 2: linux+amd+radv+dgpu AMD Radeon RX 9070 (RADV GFX1201), 26.0.8 AMD Ryzen 7 5800X 8-Core Processor, 32 GB RAM
Scene: field-default, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6, quality_preset=3); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); field-default: 2 game starts a side, field 30 s after 5 s warm-up; timed builds: main and this PR each built 2 times a cache, alternating, each a whole build in a fresh folder (lift; shader preparation not timed); each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (@Beanstalk on this PR runs it again)

Build times (each a median of the timed builds; caches: lift = the lift again (no kept lift, main's compiled objects kept))

Before: main @ 3df5c13 After: this PR @ 8083880
lift · lifted C main's not compared

Flags and settings

None: the plan varied no setting, switch or build option.

As Markdown for the PR description:

## Measurements

**Verdict:** could not be judged on linux (2 reports).

Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM`; 2: linux+amd+radv+dgpu `AMD Radeon RX 9070 (RADV GFX1201), 26.0.8` `AMD Ryzen 7 5800X 8-Core Processor, 32 GB RAM`  
Scene: field-default, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6, quality_preset=3); each cell's settings and switches as the table names them  
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); field-default: 2 game starts a side, field 30 s after 5 s warm-up; timed builds: main and this PR each built 2 times a cache, alternating, each a whole build in a fresh folder (lift; shader preparation not timed); each number the median of the runs, then of 2 reports  
How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again)

Build times (each a median of the timed builds; caches: lift = the lift again (no kept lift, main's compiled objects kept))

| | Before: main @ 3df5c13 | After: this PR @ 8083880 |
|---|---|---|
| lift · lifted C | main's | not compared |

**Flags and settings**

None: the plan varied no setting, switch or build option.
The testers' hardware
Target Report Ran on
linux 1 linux+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 5070, 620.70.6 AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM
linux 2 linux+amd+radv+dgpu AMD Radeon RX 9070 (RADV GFX1201), 26.0.8 AMD Ryzen 7 5800X 8-Core Processor, 32 GB RAM
<!-- beanstalk:previous --> <details><summary>Beanstalk test farm · an earlier run: commit <code>8083880</code> (the newer results are in a comment below)</summary> Commit `8083880` · approved by futondestroyer (reviewed) · [dashboard](https://beanstalk.magicbean.dev/prs/1929) ⚠️ **Note:** The plan compares the PR build with main; the schema cannot pin the requested head and merge-base hashes. ⚠️ **Note:** A candidate-only Sprint-on scenario is not supported: Sprint is not an available plan setting, and scenarios cannot run only on the PR. Physical controls, rebinding across restart, and short physical A taps are also outside this plan. ⚠️ **Note:** The frame-time segment is stationary after the scripted inputs; the pictures compare their outcomes, but cannot prove a roll or movement speed on their own. Started at 2026-11-20 23:07 UTC. **Estimate:** done in about 1 h 15 min, from the queue and the testers online now. | Target | Result | Reports | Frame time p99 (ms), main → PR | VRAM peak (MB) | Crashes | | --- | --- | --- | --- | --- | --- | | `linux` | ⚠️ couldn't be judged | 2 [1](https://beanstalk.magicbean.dev/reports/rp_d91ee73444b23f41) [2](https://beanstalk.magicbean.dev/reports/rp_84635f8b7b646387) | — | — | none | | `windows` | 🔄 running | 0 of 1 | — | — | — | **Couldn't be judged**: the tester's PC could not finish the test. The farm tries another PC, or the same one after a fix; it says nothing about the PR on its own. - `linux`, [report 1](https://beanstalk.magicbean.dev/reports/rp_d91ee73444b23f41): `the main base failed (exit 1): build.py stopped (exit status 1): <data>/longshot/nx2/../gpu/vk/vk_vma.cpp:421:9: error: return type 'unsigned long' must match previous return type 'unsigned long long' when lambda expression has unspecified ` - `linux`, [report 2](https://beanstalk.magicbean.dev/reports/rp_84635f8b7b646387): `the main base failed (exit 1): build.py stopped (exit status 1): <data>/longshot/nx2/../gpu/vk/vk_vma.cpp:421:9: error: return type 'unsigned long' must match previous return type 'unsigned long long' when lambda expression has unspecified ` <details><summary>What was tested</summary> **What will be tested** · AI plan (`gpt-6.1-terra`) On Linux and Windows, both revisions are built and run twice in uncapped Hyrule Field gameplay with Sprint left off. Scripted movement and A presses produce comparison pictures, followed by measurements of median and p99 frame time and 1% lows. Crashes, device loss, significant frame-time regressions, or picture differences fail; gameplay takes about 20 minutes per target, in addition to timed lift builds. - **Where:** linux (1-2 reports), windows (1-2 reports). - **field-default:** from the repository's save `kokiri-start` (settings vsync=2, frame_rate_cap=6, quality_preset=3), then teleport to Hyrule Field, move the stick, take the picture after-movement to compare with main, press buttons, take the picture after-a to compare with main, measure field for 30 s (after 5 s warm-up); 2 runs on each side. - **Build times:** main and the PR each built 2 times, alternating, each whole in a fresh folder: the lift again (no kept lift, main's compiled objects kept); every stage timed (shader preparation is not), and the lifted C compared. - **Measured:** crashes, frame times, stutters and hitches, the game's log. - **The PR fails if:** it crashes; or a picture differs from main by more than 5%; or median frame time is more than 15% above main; or frame time (p99) is more than 15% above main; or forbidden log lines is above 0; or frame time (worst 1%) is more than 15% above main. - **Time:** about 23 min of game runs per tester (4 game starts), plus the two builds and 4 timed builds (about 80 min). </details> <details><summary>Measurements on `linux`, for the PR description</summary> **Verdict:** could not be judged on linux (2 reports). Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM`; 2: linux+amd+radv+dgpu `AMD Radeon RX 9070 (RADV GFX1201), 26.0.8` `AMD Ryzen 7 5800X 8-Core Processor, 32 GB RAM` Scene: field-default, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6, quality_preset=3); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); field-default: 2 game starts a side, field 30 s after 5 s warm-up; timed builds: main and this PR each built 2 times a cache, alternating, each a whole build in a fresh folder (lift; shader preparation not timed); each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Build times (each a median of the timed builds; caches: lift = the lift again (no kept lift, main's compiled objects kept)) | | Before: main @ 3df5c13 | After: this PR @ 8083880 | |---|---|---| | lift · lifted C | main's | not compared | **Flags and settings** None: the plan varied no setting, switch or build option. As Markdown for the PR description: ```markdown ## Measurements **Verdict:** could not be judged on linux (2 reports). Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM`; 2: linux+amd+radv+dgpu `AMD Radeon RX 9070 (RADV GFX1201), 26.0.8` `AMD Ryzen 7 5800X 8-Core Processor, 32 GB RAM` Scene: field-default, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6, quality_preset=3); each cell's settings and switches as the table names them Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); field-default: 2 game starts a side, field 30 s after 5 s warm-up; timed builds: main and this PR each built 2 times a cache, alternating, each a whole build in a fresh folder (lift; shader preparation not timed); each number the median of the runs, then of 2 reports How to reproduce: the farm's plan above (`@Beanstalk` on this PR runs it again) Build times (each a median of the timed builds; caches: lift = the lift again (no kept lift, main's compiled objects kept)) | | Before: main @ 3df5c13 | After: this PR @ 8083880 | |---|---|---| | lift · lifted C | main's | not compared | **Flags and settings** None: the plan varied no setting, switch or build option. ``` </details> <details><summary>The testers' hardware</summary> | Target | Report | Ran on | | --- | --- | --- | | `linux` | [1](https://beanstalk.magicbean.dev/reports/rp_d91ee73444b23f41) | linux+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 5070, 620.70.6` `AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAM` | | `linux` | [2](https://beanstalk.magicbean.dev/reports/rp_84635f8b7b646387) | linux+amd+radv+dgpu `AMD Radeon RX 9070 (RADV GFX1201), 26.0.8` `AMD Ryzen 7 5800X 8-Core Processor, 32 GB RAM` | </details> </details>
Owner
@fix
Member

I didn't change this pull request: what the review asks needs the author (runs, measurements or decisions, which can't be done here).

Already done

  • Controller release after menu — host/nn_hid.c:941: Sprint excludes pad_after_menu; tests/test_pads_after_menu.py:100 covers release and repress after closing the menu.
  • Restore only automatic bindings — gpu/vk/menu/sprint.cpp:25: host/nn_hid.c:660 tracks binding edits and validates provenance, gpu/vk/vk_menu.cpp:1063 saves it, and tests/test_sprint.py:92 covers restoration across restarts and player edits.
  • Preserve A when Sprint ends — host/nn_hid.c:963: The pending press survives stopping or releasing Sprint; tests/test_sprint.py:290 covers both transitions.
  • Deliver A after a long interval — host/nn_hid.c:991: The pending press is emitted before its delivery window begins; tests/test_sprint.py:301 covers a long interval between reads.

Left for the author

  • Measure Sprint performance: Measure frame times and lows on the current head against the base in a representative uncapped or heavy scene; capped FPS does not establish the per-frame cost.
  • Physical short A taps: Exercise short A taps with physical keyboard and controller input, including at low frame rates, to verify the delayed press in play.
  • Current-head platform and gameplay verification: Run the current head on Linux and Steam Deck and with physical controls; cover sword use, boots, Epona and swimming. Obtain the Linux build, Tier 1 and current-head CI results.
  • Coordinate with #196: Confirm the status of #196 with its author and coordinate the merge order so one implementation is retained.

Nothing was pushed.


Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.

I didn't change this pull request: what the review asks needs the author (runs, measurements or decisions, which can't be done here). **Already done** - **Controller release after menu — host/nn_hid.c:941**: Sprint excludes pad_after_menu; tests/test_pads_after_menu.py:100 covers release and repress after closing the menu. - **Restore only automatic bindings — gpu/vk/menu/sprint.cpp:25**: host/nn_hid.c:660 tracks binding edits and validates provenance, gpu/vk/vk_menu.cpp:1063 saves it, and tests/test_sprint.py:92 covers restoration across restarts and player edits. - **Preserve A when Sprint ends — host/nn_hid.c:963**: The pending press survives stopping or releasing Sprint; tests/test_sprint.py:290 covers both transitions. - **Deliver A after a long interval — host/nn_hid.c:991**: The pending press is emitted before its delivery window begins; tests/test_sprint.py:301 covers a long interval between reads. **Left for the author** - **Measure Sprint performance**: Measure frame times and lows on the current head against the base in a representative uncapped or heavy scene; capped FPS does not establish the per-frame cost. - **Physical short A taps**: Exercise short A taps with physical keyboard and controller input, including at low frame rates, to verify the delayed press in play. - **Current-head platform and gameplay verification**: Run the current head on Linux and Steam Deck and with physical controls; cover sword use, boots, Epona and swimming. Obtain the Linux build, Tier 1 and current-head CI results. - **Coordinate with #196**: Confirm the status of #196 with its author and coordinate the merge order so one implementation is retained. Nothing was pushed. --- <sub>Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.</sub>
Owner
@fix
llun force-pushed feat/sprint from 8083880e5d
Some checks reported errors
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 10s
PR description / Description follows the template (pull_request_target) Failing after 17s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 22s
CI / Workflow lint (actionlint) (pull_request) Successful in 23s
Third-party parts / As their notes say (pull_request) Successful in 13s
Website / Website on Cloudflare (pull_request_target) Has been skipped
beanstalk/approval approved by futondestroyer (reviewed)
beanstalk/linux 0/1 pass, 2 error
beanstalk/windows 0/1 pass, 2 running, 0 queued
CI / Website build and checks (pull_request) Has been cancelled
to 297f977dd9
Some checks failed
beanstalk/approval not approved yet
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 15s
PR description / Description follows the template (pull_request_target) Failing after 19s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 18s
CI / Workflow lint (actionlint) (pull_request) Successful in 21s
Third-party parts / As their notes say (pull_request) Successful in 22s
CI / Website build and checks (pull_request) Has been cancelled
2026-11-21 00:04:52 +00:00
Compare
Member

Beanstalk test farm

Commit 2bfa4b2 · approved by nobody yet · dashboard

A member of BigBeans can review this commit and comment @Beanstalk to test it.

<!-- beanstalk:sticky --> ### Beanstalk test farm Commit `2bfa4b2` · approved by nobody yet · [dashboard](https://beanstalk.magicbean.dev/prs/1929) A member of `BigBeans` can review this commit and comment `@Beanstalk` to test it.
Member

I didn't fix this pull request: the push was refused (error: failed to push some refs to 'http://magicbean-git:3000/llun/Longshot.git'). Nothing was pushed.


Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.

I didn't fix this pull request: the push was refused (error: failed to push some refs to 'http://magicbean-git:3000/llun/Longshot.git'). Nothing was pushed. --- <sub>Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.</sub>
fix(input): Sprint off puts B back on Shift when the player rebound only Sprint's key
Some checks failed
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 8s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 13s
CI / Workflow lint (actionlint) (pull_request) Successful in 14s
Third-party parts / As their notes say (pull_request) Successful in 18s
PR description / Description follows the template (pull_request_target) Failing after 7s
beanstalk/approval not approved yet
CI / Website build and checks (pull_request) Failing after 17s
2bfa4b2293
Found in the game on 297f977d: with the row on, Sprint rebound to G (key_sprint=71) and the game started again, turning
the row off kept G on Sprint and put L3 back on LS, but left B on V and Shift unused. The provenance dropped B's
automatic move together with Shift's whenever Sprint's key changed, though the player never touched B. B's move now
stands on its own: it is restored while B is still on V and no control the player chose has Shift (Sprint's own
automatic Shift aside), so it never takes Shift from a control the player gave it to. tests/test_sprint.py:
rebound-sprint (fails before: B stays on V) and shift-taken (Shift given to X: B stays on V).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Owner
@fix
Author
Member

Rebased onto main @ dda8b119 (297f977d), then 2bfa4b22. What changed and what ran:

  • The reviewers' changes are in and checked (BeanBot's, asked for by HumanBean, and futondestroyer's): the menu-release guard on Sprint's pad read, the A tap delivered when Sprint ends or a frame runs long, and the saved ownership of the bindings the row moves (sprint_auto_bind). I read each against the code; they hold.
  • In the game on 297f977d (Windows 11, RTX 5080, headless): Sprint's speeds as before (child Sprint 5.28-5.37 m/s against A held 5.21-5.35 and the plain run 4.44-4.5; adult 6.34 against 6.3, row off 5.25-5.36). Across a restart (the bridge's relaunch): row on, restart, row off put B back on Shift and L3 back on LS, Sprint with nothing, sprint_auto_bind gone. So the point in #1929 (comment) and #issuecomment-60820 (the moves forgotten at a restart) no longer holds.
  • One case it didn't cover, fixed in 2bfa4b22: with Sprint rebound to G before a start, row off kept G and put L3 back but left B on V, Shift unused. B's move now stands on its own (back to Shift while B is on V and no control the player chose has Shift). tests/test_sprint.py has the case (fails on 297f977d) and one where Shift was given to X (B stays on V). The in-game rerun on 2bfa4b22 follows.
  • The whole suite on 297f977d against main @ dda8b119: 709 passed against 708, the same 64 failing (environment ones here).
  • Measurements: this PR against main in Hyrule Field, within the spread (median 8.30 against 8.32 ms, p99 11.01 against 10.87, 1% low 90.9 against 92.0 fps) but held at the game's 120 fps cap; the uncapped A/B is running and goes in the next comment.
  • HumanBean's @fix at 00:03 was refused because my rebase's force-push landed at 00:04:52, just before BeanBot's push: nothing in the settings (maintainer edits are on).
Rebased onto `main` @ dda8b119 (297f977d), then 2bfa4b22. What changed and what ran: - **The reviewers' changes are in and checked** (BeanBot's, asked for by HumanBean, and futondestroyer's): the menu-release guard on Sprint's pad read, the A tap delivered when Sprint ends or a frame runs long, and the saved ownership of the bindings the row moves (`sprint_auto_bind`). I read each against the code; they hold. - **In the game on 297f977d** (Windows 11, RTX 5080, headless): Sprint's speeds as before (child Sprint 5.28-5.37 m/s against A held 5.21-5.35 and the plain run 4.44-4.5; adult 6.34 against 6.3, row off 5.25-5.36). **Across a restart** (the bridge's `relaunch`): row on, restart, row off put B back on Shift and L3 back on LS, Sprint with nothing, `sprint_auto_bind` gone. So the point in https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/1929#issuecomment-60818 and #issuecomment-60820 (the moves forgotten at a restart) no longer holds. - **One case it didn't cover, fixed in 2bfa4b22:** with Sprint rebound to G before a start, row off kept G and put L3 back but left B on V, Shift unused. B's move now stands on its own (back to Shift while B is on V and no control the player chose has Shift). `tests/test_sprint.py` has the case (fails on 297f977d) and one where Shift was given to X (B stays on V). The in-game rerun on 2bfa4b22 follows. - **The whole suite** on 297f977d against `main` @ dda8b119: 709 passed against 708, the same 64 failing (environment ones here). - **Measurements:** this PR against `main` in Hyrule Field, within the spread (median 8.30 against 8.32 ms, p99 11.01 against 10.87, 1% low 90.9 against 92.0 fps) but held at the game's 120 fps cap; the uncapped A/B is running and goes in the next comment. - HumanBean's `@fix` at 00:03 was refused because my rebase's force-push landed at 00:04:52, just before BeanBot's push: nothing in the settings (maintainer edits are on).
Member

Nothing needed changing: the review's points are already done.

Already done

  • Sprint binding provenance: The current head persists automatic binding provenance in gpu/vk/vk_menu.cpp:1063, reloads and validates it in host/nn_hid.c:609, and clears ownership after later binding edits in host/nn_hid.c:738. Restart and rebind cases are covered in tests/test_sprint.py:93. Nothing was run.

Nothing was pushed.


Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.

Nothing needed changing: the review's points are already done. **Already done** - **Sprint binding provenance**: The current head persists automatic binding provenance in gpu/vk/vk_menu.cpp:1063, reloads and validates it in host/nn_hid.c:609, and clears ownership after later binding edits in host/nn_hid.c:738. Restart and rebind cases are covered in tests/test_sprint.py:93. Nothing was run. Nothing was pushed. --- <sub>Automated fix by BeanBot (gpt-6.1-terra), asked for by HumanBean. Nothing was run; a maintainer decides.</sub>
llun changed title from feat(input): sprint on Shift or the left stick's press, the game's own post-roll run (#196 carried onto main) to WIP: feat(input): sprint on Shift or the left stick's press, the game's own post-roll run (#196 carried onto main) 2026-11-21 00:46:16 +00:00
Author
Member

Back to WIP: the description check fails on one box, the host code built on Linux. It's built and run on Windows only here; the Test Farm's Linux PCs couldn't build main itself (vk_vma.cpp:421, not this PR's). It goes ready again once a Linux build of the head is in, with the uncapped A/B and the in-game rerun on 2bfa4b22.

Back to WIP: the description check fails on one box, the host code built on Linux. It's built and run on Windows only here; the Test Farm's Linux PCs couldn't build main itself (vk_vma.cpp:421, not this PR's). It goes ready again once a Linux build of the head is in, with the uncapped A/B and the in-game rerun on 2bfa4b22.
HumanBean changed title from WIP: feat(input): sprint on Shift or the left stick's press, the game's own post-roll run (#196 carried onto main) to feat(input): sprint on Shift or the left stick's press, the game's own post-roll run (#196 carried onto main) 2026-11-21 01:05:00 +00:00
HumanBean deleted branch feat/sprint 2026-11-21 01:05:10 +00:00
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
5 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
MagicBeanDev/Longshot!1929
No description provided.