[doodlewind]

OpenStrike moves to the current PocketJS and plays the Pocket3D title card when it starts on PS Vita, PSP and Nintendo 3DS: the mark and the wordmark on the plum ground, 144 ticks at 60 Hz (2.4 s), drawn by the CPU into the display frame buffer before the renderer is initialized.

What changed

vendor/pocketjs moves 1c2002a2 → 757e90ec (five commits: a Vita GXP alignment fix, the 3d.pocket.nexus site, the pocket3d-title crate, the Pocket3D License). None of them changes a file under hosts/; crates/openstrike-3ds/host.patch applies to the same host sources as before.

Console	Call	Where	Notes
PS Vita	pocket3d_title::vita::play()	crates/openstrike-vita/src/wired.rs, first statement of run(), before graphics::init_with_pool	main() only calls wired::run(), so this is the one launch path. Dependency added under the Vita target.
PSP	title()	crates/openstrike-psp/src/main.rs, in run() after the clock is set and before host::init_graphics	Draws 8888 into the uncached mirror of video memory (sceGeEdramGetAddr() | 0x4000_0000), 480 x 272, stride 512. Zeroes those bytes when the card has ended (see below).
Nintendo 3DS	pocket3d_title_play()	crates/openstrike-3ds/host.patch, after gfxInitDefault() and before C3D_Init	The patch also adds #include <pocket3d_title.h>; crates/openstrike-3ds/Makefile adds the crate's include/ directory. The vendored host stays unmodified.

README.md gains a "Pocket3D title card" section: the credit with a link to https://3d.pocket.nexus, the three call sites, and the rule that the card is not skipped, shortened, recoloured or redrawn here. The repository has no AGENTS.md or CLAUDE.md, so the rule lives there.

PSP hand-over. The card ends on black pixels whose fourth byte is 255. With the default 16-bit frame buffer the GE's display buffer occupies the same bytes and is shown from sceGuDisplay(true) until the first frame is drawn, so it would read them as alternating 0x0000 / 0xff00 columns. title() therefore fills the bytes it used with zero before it returns.

No skip switch. None of the three hosts reads a development boot switch before graphics start, so every build plays the card, capture and benchmark builds included. Capture frame indices are counted in the host loop and are not shifted by it (checked on PSP, below).

Platforms without the card

These builds have no drawer in pocket3d-title for their display path and do not show the card yet. This PR adds none.

• Desktop (crates/openstrike, wgpu)
• iPod touch 4 (crates/openstrike-ipod)
• Nokia E7 (crates/openstrike-symbian)
• The static map viewer (site/map-viewer). pocket3d-title ships a browser playTitle(); the viewer does not call it here.

domain/crates/gu-demo and the pocket3d-vita fullscreen_sky example are renderer bring-up utilities and are unchanged.

License

The pinned PocketJS puts Pocket3D (pocket3d/, devices/, engine/pocket3d/ in vendor/pocketjs) under the Pocket3D License 1.0: MIT's grant plus the condition that a distributed product drawing 3D scenes with Pocket3D shows the title card at each start. OpenStrike's own LICENSE (MIT) is unchanged. For a platform without a drawer the license asks for the mark and wordmark from the artwork for at least two seconds (condition 2b); the four builds listed above do not do that yet.

Verification

Everything below ran on a macOS host. Nothing was run on a console.

Cooked maps (<maps>: the eight classic maps and wwdc24-parkour.p3d) were supplied from a local directory; each build verified them with pocket3d-cook --verify-cooked.

Host builds

Command	Result
bun scripts/psp.ts --release --cooked-maps <maps> --package	pass: EBOOT.PBP, 5,263,106 bytes
OPENSTRIKE_COOKED_MAPS=<maps> bun scripts/vita.ts --release	pass: dist/vita/OpenStrike.vpk, 26,949,949 bytes (USB debugging on, the default); pocket3d_title::draw, sceDisplaySetFrameBuf and sceKernelAllocMemBlock are in the ELF
OPENSTRIKE_COOKED_MAPS=<maps> bun run build:3ds	pass: dist/3ds/openstrike.3dsx, 3,636,052 bytes; pocket3d_title_draw and pocket3d_title_art_half are in the link map
OPENSTRIKE_COOKED_MAPS=<maps> bun scripts/3ds.ts --capture	pass (compile and link of the capture configuration only)
bun run build:3ds --pocket-only	pass

Tests the workflows run (Rust stable 1.98.0)

CI on this PR (commit 518d186): all seven jobs pass, including bun run test:3ds with the lease test and the iPod cargo miri test job. The table is the local run.

Workflow	Command	Result
simulation	cargo test --locked -p openstrike-core -p openstrike-character -p openstrike-mods -p openstrike-companion	pass
simulation	cargo check --locked -p openstrike-core --no-default-features --features libm	pass
simulation	cargo check --locked --manifest-path crates/openstrike-symbian/Cargo.toml --lib	pass
simulation	cargo test --locked --manifest-path crates/openstrike-vita/Cargo.toml --lib	pass
simulation	bun test test/strike-sdk.test.ts test/openstrike-vita-artwork.test.ts	pass (3)
domain	cargo test --locked --manifest-path domain/Cargo.toml	pass
domain	cargo check --locked --manifest-path domain/Cargo.toml -p pocket3d-bsp --no-default-features --features libm	pass
domain	cargo test --locked --manifest-path domain/crates/pocket3d-vita/Cargo.toml	pass
domain	cargo test --locked --manifest-path domain/crates/pocket3d-gles2/Cargo.toml	pass
domain	cargo check --locked -p openstrike	pass
domain	bun test test/cook-map.test.ts test/symbian-tooling.test.ts test/psp-toolchain.test.ts	pass (15)
ipod	bun run test:ipod	pass
ipod	cargo check --locked --manifest-path crates/openstrike-ipod/Cargo.toml --features native-check	pass
3ds, ipod	bun run check:platforms (includes bun run typecheck)	pass
3ds	bun run test:3ds	1 test failed locally: test/3ds-device-lease.test.ts reported Device busy: 3ds:wire, because that lease was held by another process on the machine. It was not retried locally; it passes in CI.
3ds	the rest of test:3ds, run separately: python3 -B -m unittest discover -s test -p test_3ds_deploy.py; bun test test/3ds-tooling.test.ts test/tactical-touch.test.ts; cargo test --locked --manifest-path crates/openstrike-3ds/Cargo.toml	pass (4; 11; 9)
psp-map-read	bun test test/psp-map-read.test.ts test/psp-memory.test.ts	pass (4)

test/3ds-tooling.test.ts applies host.patch to the pinned host, so it covers the extended patch.

Emulator: PPSSPPHeadless, software renderer, at the commit in test/goldens-psp/PPSSPP-COMMIT.txt

Check	Result
OPENSTRIKE_COOKED_MAPS=<maps> bun scripts/e2e-psp.ts, liveness	pass: the capture EBOOT plays the card, boots and dumps every frame of the four cases (4, 44, 12 and 30 frames)
Same run, golden comparison	fails, 8 of 8 shots, with and without this change. The goldens were last recorded on 2026-09-18 with the 32-bit frame buffer; scripts/psp.ts has defaulted to the 16-bit buffer since 4b528d5 (2026-09-20). A captured frame has 611 colours, its golden 12,247.
Frames with the card against frames of the same tree with the title(); call removed	byte-identical, 8 of 8 shots, in the default 16-bit build and in a --framebuffer32 build. The 32-bit frames are within 2 to 405 of 130,560 pixels of the goldens in both.
The card itself, from a temporary instrumented build (not committed) that dumps video memory at tick 60	FNV-1a 64 of the 512 x 272 x 4 bytes is 0x6605dcb9c1790b74, the PSP frame hash recorded in pocket3d-title's tests. The display frame buffer reads back as 0x44000000, width 512, format 8888, and the present closure ran 144 times.

Not done

• Not run on a PSP, a PS Vita or a Nintendo 3DS. The fade timing, the hand-over to GXM / the GE / citro3d and the first frames after the card are unobserved on hardware. The PSP zero fill is reasoned from the buffer layout and the dumped end-of-card bytes, not seen on a screen.
• bun scripts/e2e-vita.ts (Vita3K) and bun run test:e2e:3ds (Azahar) were not run; both open an emulator window. Their capture builds now play the card before frame 0.
• The cargo miri test job of the iPod workflow was not run locally (it passes in CI).
• The PSP goldens were not re-baselined.

🤖 Generated with Claude Code