summaryrefslogtreecommitdiff
path: root/src/host
Commit message (Collapse)AuthorAge
* audio: a GPU buffer streamed within the frame, at a lead the host measuresGabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | A compute shader's output was heard only after the sketch read it back (f.download, a frame later) and pushed it behind a lead of its own choosing; spheres kept 100 ms, and a change it made left for the device 96 ms later on average (measured, dummy driver, 60 Hz). f.audioBuffer(key, buffer, .{ .count, .offset }) streams `count` f32 a shader wrote: the host copies them back while it runs the frame's commands, waiting on the copy's fence (Executor.read: 0.2 ms on average here, 2 ms at the 99th percentile on a loaded machine), and queues them on the stream, so nothing is in flight between frames and the sketch keeps no books. Input.lead says how far ahead to make sound: the slowest of the last 64 frames' worth of samples (timed between the ends of the frames' commands, where the pushes are), plus the device's pull (what SDL takes from a stream at once, read from the device: 1024 here), plus half a pull for each underrun, a sample less each frame after; capped at 8192. A sketch pushes `in.lead -| in.stream(key).queued` a frame, and its change leaves after about what is queued ahead of it, a pull: 19 to 27 ms on average with SDL's 1024-sample pulls, 6.3 to 6.5 ms with 256 (SDL_AUDIO_DEVICE_SAMPLE_FRAMES; 17 under heavy load). Stream.underruns counts the pushes that found a stream empty though the frame before had fed it (a gap). The voices keep their 2048 samples: mixed to the lead, a one-shot longer than it is mixed over several frames, and voices.script's expect-audio and expect-no-audio, which look one frame at a time, failed. On the web, audio.buffer reads the buffer back with mapAsync and queues it on arrival, in order, counting it as queued meanwhile; in.lead stays at 100 ms there. The measurements (a probe patched into the host for the purpose, kept out of the tree) are in the 02-pardes-code scratch, .scratch/sketch/gpusynth. Tests: the lead's arithmetic (audio.zig, in test-audio's filter); spheres' play script checks its stream is fed. zig build test: 164/164 steps, 49/49 tests run (the rest cached); zig build test-audio: 48/48 steps, with the voices sketch's wasm, so web.zig compiles; voices.script passes four runs in four. web/host.js: node --check only, no browser run. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* Merge the audio work and the incremental reload loopGabriel Schneider3 days
|\ | | | | | | | | | | | | The mixer's knee, newSound, sfxr, voice rate and groups, the scope's public pieces, micscope and spheres as examples; with the loader workaround for Zig 0.17.1's incremental linker and the panic-trace link. The README's hot-reload bullet takes the incremental loop's paragraph; zig build test: 164/164 steps, 132/132 tests. Co-Authored-By: Claude Fable 5.2 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
| * Keep a loaded sketch library's name valid, for panic tracesGabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | A panic's stack trace finds a library's debug info by the name dlopen was given, and the load link was deleted right after loading: every frame in a sketch printed as ??? (a sketch that links libc; one that does not is not seen by Zig's own dl_iterate_phdr at all). Link and fd now live as long as the library is loaded, and Library.close removes them; a host that dies leaves its dangling links in $XDG_RUNTIME_DIR. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
| * Load libraries from Zig 0.17.1's incremental linkerGabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | `-fincremental` selects Zig's new ELF linker, and 0.17.1's writes DT_VERSYM without DT_VERDEF or DT_VERNEED when nothing is versioned: glibc then reads a null version table at the first relocation against an export, and dlopen faults. When a library has that shape, the host loads a copy of it in an anonymous memory file with DT_VERSYM dropped (nothing in it is versioned, so nothing is lost); the build's own file is never written, and every other library loads as before. The copy is 10 ms for the counter's 28 MB library. Fixed in Zig after 0.17.1 (6d27011edb): the check then never fires. The reload test's second build goes through the new linker. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* | audio: a voice's rate (its pitch), and groups of voices under one gainGabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Voice.rate is the voice's playback speed and so its pitch: 2 is an octave up and half as long, 0 holds it, and changing it does not restart the voice. The mixer keeps a fractional cursor and resamples by linear interpolation between a sample and the next (the last toward silence, or the first when looping): the least code that sounds right, a little dull far from 1; at rate 1 the samples come out exactly. An infinite or NaN rate ends the voice instead of crashing the conversion. Voice.group (0-7) puts a voice under a group's gain, which draw declares each frame with f.gain(group, value) (Command.gain); a group not given one plays at 1, so it is declared state and replays like the voices. The mixer glides each group toward its gain by 1/480 a sample (10 ms for a full swing), so a gain set in one step does not click. Music ducking under speech is then the sketch lowering one number; nothing ducks by itself. The web host ignores gains as it ignores voices. Test: half and double speed against the interpolated samples, rate 0 holding, an infinite rate ending; a looping voice in group 1 at gain 0.5 gliding there and staying while group 0's is untouched, and going back to 1 when the gain is no longer declared. test-audio also builds the voices sketch for the web, which compiles the web host's newSound and gain. The README lists rate, group and f.gain with what they do. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* | host.newSound: samples a sketch made become a sound, played by voicesGabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | A sketch that synthesised a one-shot had to stream it (and keep a replay from re-firing it by hand) or encode a WAV for decodeSound. host.newSound (samples) copies mono samples at in.sample_rate into the host and gives an Audio like a loaded one, so the one-shot is a voice: reconciled by key and serial, mixed, faded and limited by the host. It is kept for the life of the host like every sound, and .none if a sample is not finite. The vtable entry is C callconv (pointer and length), as anything new in abi.zig is, for a second host; the web host keeps the copy for samples() (its voices are silent). Tests: a unit test that the sound is a copy and that NaN and infinity are refused; test/voices.zig makes a 1000-sample blip in setup and declares it under a serial space bumps, and test/voices.script checks it sounds once, stays silent through a recording of it played, stepped and seeked, and sounds again for a new serial. For that, expect-audio KEY also passes when voice KEY was mixed since it was last asked, and expect-no-audio KEY says it was not. zig build test-audio runs the script too. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* | audio: the voice mix plays at true volume below a knee and is limited only ↵Gabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | above it Every mixed sample went through x/(1 + |x|), so one voice at 0.5 played at 0.33, bent. The mix now passes through untouched up to 0.75 (-2.5 dBFS) and above it follows knee + (1 - knee)·u/(1 + u), u the excess over the room left (the old curve is its knee-0 case): same value and slope at the knee, rising, tending to 1. A clip normalised to full scale only has its peaks bent (1 plays at 0.875); two at full scale sum to 0.958. Per sample, no state, no lookahead, on the frame thread as before. A unit test checks the curve: identity to the bit below the knee (every 997th float, both signs), rising and at most 1 from the knee to infinity (every float near the knee), odd, slope 1 at the knee; `zig build test-audio` runs the mixer's tests alone. The README says what the mixer does, and that SDL clips the total of the voice mix and the streams. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* | Bound a scripted run's memory sessions; check a script records the same twiceGabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | A scripted run from fresh state records in memory, and a very long one grew without end. It stops at 256 MiB (ten hours or more of play at 60 Hz; a write the memory database cannot make, out of memory included, is the same): once, a warning, and no more frames, while what it holds still reads (marks, clips, past sessions). test-same plays test/past.script twice into two databases and compares them value by value with test/same_sessions.zig, but for a session's started and ended (the wall clock): every frame, keyframe, stretch and mark follows from the script alone. A third run from fresh state leaves its state file and no database. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* | Mend a sessions database a crash tore, compact one at exitGabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | A database whose last commit a power loss tore (its root header written, its bytes not all) was refused and renamed aside, a new one begun. Now it is opened at the last commit before the tear whose trailer verifies and whose sessions read whole, scanned back from the end, as pardes's state file is; renaming it to <path>.broken.<seconds> stays the last resort, when no commit is whole, and nothing is deleted. Every commit leaves the nodes it copied behind, about a kilobyte a second of play: a clean exit compacts the file (into <path>.new, synced, renamed over it) once it is 64 MiB and past twice what its batches and keyframes hold plus 64 MiB. Tests: a database of two sessions, cut at six places inside its last commit, opens at the commit before with every frame of both; one with no whole commit is kept aside byte for byte; a compacted one is smaller and holds every frame and keyframe. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* | Export a session to a file of its own, and import oneGabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | export-session [S] FILE writes session S (this run's by default, committed first), with the marks and clips that point into it, as session 0 of a new database: written beside FILE, synced, renamed over it. import FILE appends each session of such a file to this database, from "import <file name>", its marks and clips renamed NAME@<file name> (cut to 32 bytes). The copy is value by value, the same few lines for both, over whatever a session holds; sessions lists where one came from. test-import records test/past.script's session, exports it, imports it into another database and replays it there: the mark it carries is the State half a second in exactly, and two seconds in is frame 120's. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* | Keep a sketch's files in .sketchbook/, a shipped game's in XDGGabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | | | | | zig build run, probe and export pass --state .sketchbook/<name>.state. The host makes that directory with a .gitignore of * (a sessions database grows, and the jj snapshot build_info takes would commit it while under a megabyte), and moves the state and store an older build kept beside it in once, with renameat2(NOREPLACE): when both exist nothing moves and the host says so. Without --state an interactive run keeps its files in $XDG_DATA_HOME/<name>/ (~/.local/share without it), not the cwd; a script or an export still uses sketch.state. test-paths checks all three cases. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* | Play past sessions back, re-synced at their keyframesGabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | play session S [at SECONDS] restores the session's keyframe before that time (a binary search over its keyframes' times), replays its frames up to it and leaves it paused there; without a time it plays from the start. At each later keyframe the replay reaches, the State is compared with it over the leaves migration kept: a keyframe taken of what update made (periodic, mark, before a reload, exit) that differs is reported once, "diverged before frame F: N values, first <field>", and every keyframe is restored, so a replay across a jump the log does not hold (set, load, a sketch command) goes on from the State that followed it. A session built from another source or build says so. The console's status line names the session, frame and time, the mark, the clip and where it records. test-past records a session, replays it by the same build (the mark's State exactly, frame 120's two seconds in, no divergence), then by a build with an edited fixture (one divergence, before the mark). Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* | Make snapshots named marks and recordings named clips of sessionsGabriel Schneider3 days
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The state file keeps live state only (SKETCH03, one slot). save NAME is a keyframe and a mark in the sessions database, synced to disk; load, diff, snapshot and expect-snapshot take mark names, through the keyframe's own layout migrated to today's, compared over the leaves migration kept. A file from before (SKETCH01/02) has its used snapshot slots imported as marks "1".."7" at open, then drops them; a run that records nothing leaves them alone. record NAME, stop and play NAME make and play clips: a frame range of a session and its start keyframe, nothing overwritten, nothing invalidated by a later save, no capacity. The 64 MiB .inputs file, its --fresh memset and Timeline's Bank are gone; numbers still work as names, so every existing script runs unchanged. Event times in a clip are the session's, not restarted at record. reload.script also checks that a field a reload drops is still in the keyframe taken before it (expect-key). Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* | Record every run's frames to a sessions database in xitdbGabriel Schneider3 days
|/ | | | | | | | | | | | | | | | | | | | | Live play was recorded only after someone typed record N, into 16 fixed slots. Now every run is a session in one xitdb database per sketch, beside its state file (<name>.xitdb): an append-only log of the frames live update was fed, one commit a second of session time, and keyframes of the State at start, every minute, on jumps, before and after reloads and at exit. A kill -9 loses at most the uncommitted second (test-kill). Whether a build records travels in the sketch: App.Options.record, null meaning Debug, reaching the host as Info.record. Scripted runs from fresh state record in memory, exports not at all. Each stretch of a session keeps its build, layout and Info.source: the repository revision (one jj or git call per build, build_info), zig version, backend and mode. Sessions.zig takes plain data (bytes, events, strings, a path, an Io), so a second host can drive it. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
* Load sketch libraries through a symlink and identify builds by statGabriel Schneider3 days
| | | | | | | | | | A load copied the whole library to $XDG_RUNTIME_DIR (a read/write loop across filesystems) and then hashed all of it for the build id: 40 ms for a 10 MB Debug library, 200 ms for drat-meout's 58 MB. dlopen only needs a fresh name, so the host links one to an fd open on the library, which also pins the file a build may rename over meanwhile. The build id recordings carry is a hash of the file's inode, size and modification time: Zig's own linker writes no build-id note in Debug.
* Refuse a sketch library built by another compiler, backend or modeGabriel Schneider3 days
| | | | | | | | abi.hash now covers the zig version, codegen backend and optimize mode: Zig's own calling convention (Host.VTable, every Allocator) differs between them, so a Debug host loading a ReleaseFast or LLVM library faulted on its first draw. Libraries export sketch_build, and the host's refusal says what differs and what to run instead.
* gpu: add a post-process pass with optional depth sampling, and describe the ↵Gabriel Schneider3 days
| | | | | | | | | | executor pardes shares `f.postProcess(source, target, .{...})` records a fullscreen pass that samples a completed render into a distinct target image: colour at fragment binding 0 and, with `sample_depth`, depth at 1, which needs the source's latest pass to have depth and one sample (missing depth, multisampled depth and feedback loops are rejected). The scene format's version 4 carries it, the executor records it on the caller's command buffer and owns the sampler, the web host does the same with its texture and sampler pairs, and native profiling reports the work as `postprocess`. The README says how to use it. docs/pardes-rendering.md describes the executor as pardes uses it for addon scenes: a borrowed SDL GPU device, `record` into the caller's command buffer, `run`, `output` and `postProcess`. The working copy as it stood on 2026-11-16; message written by Claude from the diff, at the author's request.
* Add display-paced frame timing and presentation feedback in ZigGabriel Schneider3 days
|
* Add glTF scene rendering, resource setup and 9P controlsGabriel Schneider4 days
|
* host: add optional Tracy profiling and reduce upload allocation stallsGabriel Schneider7 days
|
* Add reusable GPU backdrop filtering to the shared canvasGabriel Schneider7 days
| | | | Draw native GPU frames through the same bounded canvas used by headless capture. Backdrop commands reuse one snapshot and shader state, with an unchanged fallback on unsupported renderers. Verified through native debugger lenses, restart, and repeated font zoom checks.
* sketchbook: text rasterized at one scale and drawn at another ↵Gabriel Schneider8 days
| | | | (f.textScaled), steady under a moving camera
* sketchbook: text in six faces of the Go family, all bundled (f.textIn, ↵Gabriel Schneider8 days
| | | | host.textSizeIn, sk.Face)
* sketchbook: shapes antialiased up to 3840x2160; SDL's Vulkan backend no ↵Gabriel Schneider8 days
| | | | | | longer defragments (it lost a frame's first draws) Co-Authored-By: Claude Fable 5.2 <[email protected]>
* sketchbook: frames for ffmpeg (--export, on a clock of its own), text sizes, ↵Gabriel Schneider8 days
| | | | | | a font from the environment Co-Authored-By: Claude Fable 5.2 <[email protected]>
* sketchbook: drop a patch leftover (standalone repo: tests pass as they are)Gabriel Schneider8 days
| | | | Co-Authored-By: Claude Fable 5.2 <[email protected]>
* sketchbook: a drawing layer on top, host fixes for replay and migrationGabriel Schneider8 days
|
* sketchbook: Zig 0.17, a Store on disk, requests, a scope (waveform, ↵Gabriel Schneider8 days
| | | | spectrum, spectrogram), audio codecs and fonts, web host
* sketchbook: await in scriptsGabriel Schneider2026-09-25
| | | | | | await FIELD VALUE plays frames until a state field holds the value, for up to a minute: for waits whose length is the sketch's business, such as a line of dialogue typing out.
* sketchbook: fills blendGabriel Schneider2026-09-25
| | | | | | The renderer's draw blend mode was left at SDL's default, none, so every translucent fill (panels, labels, HUD bars) drew opaque natively while the web honored alpha. It blends now.
* sketchbook: what the game's review neededGabriel Schneider2026-09-25
| | | | | | | | | | | - input.started: true on the first update of a run, never after a reload, for a sketch to stop what should not resume from saved state. - A console line that errors fails its script. Sketch commands answer unknown, done or failed (abi.Outcome) instead of a bool that hid errors. - addScript also takes the hot-reloadable library, run through the generic host, so scripts can reload a real build. - Sounds that do not load are .none, not an empty clip. - sk_tag reads an enum field's tag name from the web build, for tests.
* sketchbook: headless runs never open a windowGabriel Schneider2026-09-25
| | | | | | | | The offscreen hints lost to the environment: with SDL_VIDEO_DRIVER=wayland set, every headless test and script opened a real window. They are overrides now. The offscreen driver has no Vulkan surface on every driver, so a headless host's GPU renderer gets no window and draws into a canvas-sized texture, which screenshots read.
* Put the game first: lib/sketchbook, sketches/vecanimGabriel Schneider2026-09-25
The root is the game. The platform library moves to lib/sketchbook (with serve_web.py, which serves any sketch's web build, as a tool of it), and vecanim to sketches/vecanim, still its own package on the same library.