Repository navigation
Conversation
…nsition, first-person)
…cale/translation, look-at z sign; reinstate + strengthen tests
Implement grid.lg: the adapter that converts xsofy's rendered glyph grid (cells with character, fg/bg colors) into vertex batches for GL rendering. Public API: - cell-quad: Convert a glyph cell to textured vertex quad (54 floats) - cell-bg-quad: Convert cell background to untextured quad - grid-verts: Batch cells into bg and glyph vertex vectors - wall-extrude-verts: Extrude a cell into 4 CCW-wound wall faces (216 floats) Coordinate system: cell (col, row) at world x = col*cell-w, y = -row*cell-h, z = 0. Cell width defaults to 0.5, height to 1.0; walls extrude upward (+z). Colors normalized from [0-255] to [0,1]; glyphs fallback to "?" if missing. Tests (grid_test.lg): 15 tests covering float counts, vertex positions/UVs/colors, alpha-zero filtering, fallback logic, batch separation, and wall geometry.
… camera; fix wall height test axis
…able
GLYPH INVENTORY: collected from terrain, entities, items, runes (140 total glyphs).
- assets/gl/runes.txt: UTF-8 string of all glyphs (ASCII 0x20-0x7E + Unicode runes)
- Baked with lgatlas (Monaco.ttf): 128x256 texture, 16x16 grid (cell 8x16)
- assets/gl/atlas.{png,edn}: texture and UV lookup table
RENDER BRIDGE (xsofy/gl/bridge.lg):
- Reuses render.lg terrain/entity composition logic without modification
- Extracts grid cells as {:col :row :glyph :fg :bg} per cell
- Applies: lighting (light/apply-light), stains (fx/stain-tint-*), gases (fx/gas-tint)
- FOV/memory/unseen cells handled: visible cells lit, remembered cells dimmed, unseen black
MAIN LOOP (main_gl.lg):
- Init GL window via glplat, load atlas texture
- Build game state via build-game (skips title screen)
- Frame loop: PollInputEvents → parse glplat strings → dispatch actions → bridge → grid-verts → submit
- Input mapping: "key:w" → "w" → input/parse-key → :unknown | "key:up" → "[A" → :up
- Ortho projection: 0–39.5 (79*0.5) on x, 0–30 on z; draw bg (tex 0) then glyphs (atlas tex)
- Auto-exit via XSOFY_GL_AUTOEXIT env var (seconds); XSOFY_GL_DUMP flag logs first frame vertex counts
VERIFICATION:
- Test suite: 362 tests, 2880 assertions, all PASS ✓
- GL version: 2-second run, 2370 cells rendered, 14220 verts/batch, lighting visible ✓
- Batch output: bg=[0 0 0 1.0] (untextured), glyph=[uvs] in atlas
NOTES FOR NEXT WAVE (fall+first-person):
- Screenfx tint flash (if used on level 2) will be a full-screen quad post-submit
- no audio calls stubbed; sound system may need wrapping at loop level
- Wall extrusion ready in grid.lg; fall.lg morph parametrization ready
- Determinism preserved: no graphics side effects on seed/action-log
…lyph stub), read-string number parsing (env parsers silently failed -> infinite loop), os/exit after -main (GLFW keeps process alive), print instead of io flush! on *out*
Implements the centerpiece of GL frontend experiment: level-2 entry triggers
playfield-to-floor transition (2.5s morph), followed by first-person mode.
LEVEL-CHANGE DETECTION:
- Track depth (:depth in world state) before/after dispatch
- On depth increase (detected in advance-pres-state), trigger fall animation
- Presentation-side state machine: :top-down → :falling → :fp (game state untouched)
- XSOFY_GL_FORCE_FALL env hook: start animation N seconds after boot for testing
MORPH RENDERING (:falling mode, 2.5s duration):
- fall/morph provides: plane-tilt-rad, wall-height, camera {eye/target/up}, projection-blend
- Component-wise MVP lerp: ortho → perspective as projection-blend (t: 0→1)
- Plane tilt: rotate-x about player pivot, composed into model matrix
- Wall extrusion: wall-extrude-verts for all wall cells (tiles 8, 9) with dynamic height
FIRST-PERSON MODE (:fp):
- Perspective projection: ~70° FOV, camera 0.5 units above player, facing heading
- Dungeon-crawler input remapping: (up/k=forward, down/j=back, left/h=turn-left,
right/l=turn-right in place, turn changes presentation-side facing only)
- Billboards: entity glyphs as upright quads facing camera (using yaw from facing)
- Wall boxes: same extrusion as morph but fixed 1.0 unit height
- Geometry culling: only render visible cells (FOV + memory + non-black cells)
DUMP METRICS (:falling only for now; can extend to :fp):
- Per-frame at t≈0, t≈0.5, t≈1: mode, t, camera eye/target, batch vertex counts
- Sanity checks: wall-batch non-empty on maps with walls; camera eye y descends
- Exit message includes final mode
fpview.lg: new namespace for FP view logic:
- remap-fp-input: input → action + facing change (turn-only flag for presentation)
- facing-to-angle: yaw from cardinal direction (for billboards)
- fp-camera-pose: camera {eye/target/up} from player pos + heading
- entity-billboard: upright glyph quad at entity cell, colored from entity visual
- Wall tile predicate: tiles 8, 9 (can extend if needed)
Test suite: 362 tests, 2880 assertions, all pass. Game logic untouched.
…op double model-tilt, 70-degree fovy (was 7), fall.lg camera in world units, force-fall no longer re-arms t=0 every frame
…eal descend fall trigger - main_gl: advance-pres-state args were swapped (compared old>new depth) so a real stairs descent never triggered the fall; only XSOFY_GL_FORCE_FALL worked - main_gl: auto-modes (autoexplore/rest/run/auto-target) now step every 60ms between inputs, mirroring play.lg's tick loop; keypress interrupts - gl/vfx.lg: tick + overlay quads for :projectile/:bolt/:flash world vfx, lifted y=0.02 over the floor; wired into all three render modes - gl/hud.lg: screen-space HUD (status line, HP bar, last 3 log messages) drawn in a second pass with near=0 ortho so it lands at NDC depth -1 (always on top under LEQUAL); fixed nested-vector corruption in hud-verts solids batch - title.lg: show-descending-screen no-ops under XSOFY_GL_FRONTEND (set by main_gl) — it painted runes into the launching terminal mid-descent Tests: 409 tests, 2968 pass, 0 fail
…sigil overlap - assets/gl: atlas rebaked from Fairfax HD (the web shell's one true face, tools/fonts/.cache) — 159 glyphs, everything the sources emit incl. the box-drawing projectile glyphs (─│╱╲) whose absence made projectiles invisible, plus full runic coverage (Monaco lacked 33 glyphs; terminal only looked right via macOS font fallback) - tools/bake_atlas.lg: baking policy now lives in lg (layout, glyph list, EDN/PNG writes) over four minimal glplat primitives (FontLoad, FontHasGlyph, FontRasterizeCell, SaveGlyphAtlasPNG); assets/gl/fonts.edn is the font manifest; replaces let-go's cmd/lgatlas Go CLI - gl/hud.lg: spell-rune glyphs render 2 cells wide with 1-cell advance so consecutive runes overlap into a composite sigil — reproduces the native terminal look (halfwidth grid + wide fallback glyphs), which Fairfax (strictly halfwidth, measured) does not produce on its own - tools/xsofy-shell.html: regenerated by gen-terminal-font.py (font fetch) Tests: 410 tests, 2972 pass, 0 fail
- glplat/Screenshot (glReadPixels → PNG, pre-swap) + XSOFY_GL_SHOT / XSOFY_GL_SHOT_AT env hooks in main_gl so frames can be inspected headlessly - fonts.edn cell 8x16 → 12x24: Fairfax HD is a bitmap-style face with a native 12x24 grid; at 16px every stroke straddled pixels and antialiased to faint hairlines (screenshot-diagnosed unreadable HUD) - bake_atlas.lg: binarize alpha (>=96 → solid) for terminal-crisp glyphs; atlas now 256x512, zero partial-alpha pixels Verified by screenshots: top-down and mid-fall frames, HUD/messages legible
wall-extrude-verts takes :fg (0-255) and shades per face (south 1.0, north 0.8, x-faces 0.65); build-wall-batch feeds it the bridge cells' lit style colors, so extruded walls follow the palette and lighting instead of rendering fullbright white. Screenshot-verified mid-fall. Tests: 411 tests, 2980 pass, 0 fail
Unset XSOFY_GL_SHOT read back as "" (truthy), so 1s in the loop called
Screenshot("") which threw; with no catch, the process exited 0 silently
— looked like a clean quit. getenv-str helper (unset/empty → nil) now
guards all env reads; the frame loop prints exceptions instead of dying
silently; title.lg's XSOFY_GL_FRONTEND guard tests count, not nil (it was
always-on, wrongly suppressing the terminal descending screen too).
Verified: plain run alive 15s+; AUTOEXIT/SHOT/DUMP paths still work.
Tests: 411, 2980 pass, 0 fail
- main loop now drives xsofy.ui/run-loop: glplat-input-source ([:key ev], pumps GLFW) + ui/clock-source, pure gl-step fold, gl-render subscriber. Same tick-vs-input contract as the terminal: sources polled input-first so a keypress cancels auto-mode before the tick's auto-step; cosmetic ticks never enter the deterministic action stream. frame-ms 0 — vsync paces via EndFrame. - :title presentation mode (default entry): scattered scheme-tinted runes pulsing via screenfx/rune-brightness at the terminal's 120ms frame pace, title fade-in, 'Press any key', seed line; any key enters gameplay. XSOFY_GL_SKIP_TITLE=1 for scripted runs. - fall trigger also checked at key-dispatch time (depth increase), not just per-tick Screenshot-verified title; force-fall runs reach :fp and exit on time. Tests: 411, 2980 pass, 0 fail
let-go now ships poll!/offer!/alts! (nooga/let-go#194), so the chan-source adapter deferred in the #56 spec is buildable: producer goroutines push events into a channel; the source drains it non-blockingly via async/poll! (capped 16/tick like input-source). The loop contract is unchanged — sources still never block, so WASM's single thread never parks the frame loop; goroutines under Go-wasm park/wake through the JS event loop. Stale 'deferred until #194' notes updated. Tests cover empty/order/splice/ flood-cap and a real (go …) producer. Tests: 416, 2988 pass, 0 fail
Collaborator
|
We rebased this branch onto current The headline: culling cells that are neither in FOV nor remembered takes the top-down frame from ~400ms to ~18ms at seed 42, and the captured frame is byte-identical before and after. Also included are frame timing behind Independent of that PR: the test suite hangs on this branch's current base. It reaches |
mparrett
self-requested a review
July 27, 2026 05:48
Owner
|
let's undraft, we are doing it 🚀 |
Collaborator
|
@nnunley #140 had fallen 46 commits behind The branch is Three ways it could go, in the order I'd suggest:
Verified on let-go
|
| Check | GLFW | Ebitengine |
|---|---|---|
| Suite | 475 tests, 0 failures | same |
| Boot to gameplay, clean exit | yes | yes |
| Title, fall, first person | yes | renders; first person mis-sorts (no depth buffer) |
| Culled top-down frame byte-identical to pre-cull | yes | yes |
| Cells / cpu per frame, seed 42 | 2,370 → 40, ~365 → ~17 ms | 2,370 → 40, ~382 → ~17 ms |
| One move per keypress, Shift+K runs, arrows move and repeat | yes | yes |
| Glyphs upright | yes | yes |
The replay-test hang nooga saw in #180's latest CI leg on 2026-09-20 came from the old base; the suite finishes on the new one.
The two new commits
2d57261movesparse-glplat-eventout ofmain_gl.lginto a purexsofy.gl.eventsnamespace, sincemain_gl.lgruns-mainon load and nothing could require it from a test.events-testcovers the Top-down perf: cull never-seen cells (22x), plus frame timing and two input fixes #180 input fixes, which nooga noted had no test. Putting the double-move back fails it.c535ab3ignores bare modifier keys. Both backends sendkey:lshiftand the like, which dispatched as an unbound key and cleared:running, so pressing Shift mid-run stopped the run. Ebitengine also auto-repeats modifiers, so holding Shift there stopped it about 20 times a second.
Two questions
- How does a player run it? The PR body points at the
glplatbranch build from let-go#392, which is closed. Today it takes let-gomainbuilt with-tags glplat(cgo, GLFW) or-tags glplat_ebiten, thenlg main_gl.lg; releasedlgbinaries register no backend. Do you want a short "running the GL frontend" section, and should it name one backend as the default? - First person on Ebitengine. C4 in let-go's backend contract makes painter's order the portable guarantee, and first person currently leans on the GLFW depth buffer. Would you ship first person as GLFW-only for now, or hold it for a back-to-front sort?
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Experiment branch (draft): the whole game rendered through a minimal OpenGL platform layer, with entering depth 2 tipping the playfield into the floor of a 3D world and continuing in first person.
Runs on the let-go
glplatbranch build (nooga/let-go#392; async parity in nooga/let-go#391):What's in the branch
main_gl.lgon the unified poll-based event loop (xsofy.ui, Responsive UI event loop: composable, replay-safe input + cosmetic animation #56): glplat input source + clock source, pure step fold, render subscriber — same tick-vs-input contract as the terminal, so cosmetic animation never touches the deterministic (seed, action-log) stream.xsofy/gl/: mat4, grid adapter (floor quads + extruded walls tinted from the lit tile palette with per-face shading), bridge (FOV/lighting/stains/gases), fall morph, fpview, vfx overlays (projectiles/bolts/flashes), HUD (status line, HP bar, log tail; spell-rune glyphs render 2-cells-wide with 1-cell advance so inscriptions overlap into composite sigils, matching the native-terminal look).tools/bake_atlas.lg+assets/gl/fonts.ednbake Fairfax HD (the web shell's single face) at its native 12x24 grid with binarized alpha;ui/chan-sourceadded for push-based producers now that let-go has poll!.Suite green throughout (416 tests, 2988 pass). Known gaps: no browser/WebGL backend yet, screenfx tint flash stub, no smooth FP camera lerp, no FOV culling of submitted geometry.