Repository navigation
0.5.3: SatDiff debut, UI pass, adjustable compressor plot, suite downloads - #2
Merged
Merged
Conversation
…ain + gain match + optional true-peak limiter + bypass/delta, no compressor, no EQ. Always-reported limiter latency, 9 tests (delayed-copy null, bypass at latency, limiter only when on, delta, sat match, gain match, telemetry) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…r input again) so the shared scope view reads it; no clipper layer Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…LAP com.thomchantra.satdiff, 13 generated params (SatComp sat slots + OUT GAIN, GAIN MATCH, LIMIT, bypass) from tools/gen_dparams.py, presets *.satdiff, editor derived from SatLim's with the limiter column replaced by GR / PEAK / RMS, LIMIT, GAIN MATCH and OUT GAIN. No factory presets yet (INIT only), no manual (alpha). 23 plug-in tests Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… unreleased); gen_lparams.py is noted as behind the committed SatLim params.rs Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…alf / from_odd_half); addition only, SatComp / SatXComp / SatLim untouched Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… no DSP behind them yet (phase, fx on / type / pre-post / freq / Q / stages / mix, tile, morph direction / depth / speed, scene A/B); engine adds input peak + mono tap telemetry for the new meters and viz; window base size 372 x 688 Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… + BYPASS, preset row with red-on-hover INIT, A|B + shape selector + LIMIT (shows GR), 275 px plot with the draw view (pencil / odd / even / exit icons, LINEAR | CURVE, accent-dashed derived half, DRAW label), hatched input meter + output meter, DRIVE / ASYM / HPF-LPF / BIAS + tile icons / TONE, MORPH block, fx row (enable, dropdown on right-click, PRE | POST) with diffusor group-delay viz, PHASE / SAT MIX, FREQ-Q / STAGES-D.MIX / OUT GAIN, vertical spectrum / oscilloscope viz. Hover effect on every toggle. Smoke tests moved to the new geometry Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…Lim (340 x 532), SatLim's 38 px cells and tighter rows, 250 px square plot with a 22 px inset on all sides, boxed square tool buttons (16 px: pencil, odd, even, close) and a bigger LINEAR | CURVE switcher, scaling dropdown next to BYPASS, INIT button retired for an oversample dropdown (1x to 8x), MORPH moved up beside the plot with the viz filling the rest of the column, input / output meters moved down beside the cell rows. Scope viz: windows 250 ms to 4 s and a retrigger (each sweep starts on the next transient, free-runs after one window) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… curve inverted the axis: the real point now follows the pointer from either side; regression test); panel narrowed to 316 x 508: plot 226 px square and three 72 px cell columns, the narrowest where the longest label + value pairs still fit (FREQ shows 20.0k above 10 kHz), BYPASS is the limiter column's width; scope: faster windows are back (5 ms to 4 s), under 500 ms the trace scrolls like a normal scope (newest at the top), from 500 ms the retriggered hard wipe; right-click flips spectrum / scope instead of a menu Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ctive Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…quadrature sin t from Niemitalo's two 4-section allpass chains, the in-phase chain a sample late; zero latency, unity gain, exact bypass at 0 degrees with a 20 ms engage fade) and the diffusor (cascade of 1 to 16 RBJ allpass stages with smoothed FREQ / Q, fx mix, enable fade) in the engine as PHASE > [diffusor PRE] > SAT > [diffusor POST] > OUT GAIN. Other fx types act as off. 14 new tests: quadrature error under 2.5 degrees from 60 Hz to 15 kHz, rotation angle within 3.5 degrees at unity gain, flat magnitude of the cascade (incl. 20 Hz / 20 kHz, Q 8, 16 stages, 44.1 / 96 kHz), stability at the extremes, impulse spread grows with stages and Q, 50 % mix notch on FREQ, bypass exactness, no added latency, PRE vs POST equal on a linear path and different through a hot shaper, phase reshapes the clipped output Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…Stage k of n now sits at FREQ * 2^(span * (k / (n - 1) - 1/2)) with span = SPREAD * 8 octaves, so the delay reaches the mids and highs instead of piling up in one band that only lags the bass (bounces: 100 stacked stages at Q 7 left a harmonic saw practically unscrambled; 40 spread stages scramble the harmonic phases over 1.5x more). Stage cap 16 -> 100 with coefficients redesigned only while FREQ / Q / SPREAD / STAGES actually move. Plug-in: STAGES is a 0..100 % param (0 % = 1 stage, 100 % = 100, default 20 %), new FX Spread param (slot 43, default 50 %), FX Mix is now Color Mix (same id), panel height 508 -> 550. 4 new tests: percentage and centre maths, flat magnitude at 100 stages incl. full spread at Q 8, spread vs stacked harmonic phase scrambling, no redesign once settled Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ble + type dropdown), then the viewport, stacked PRE-SAT / POST-SAT buttons and PHASE, then a three-slot grid (FREQ | Q, STAGES | SPREAD, free slot) with COLOR MIX, SAT MIX and OUT GAIN; panel 316 x 550, meters stretched to the new bottom. Fx viewport: faint spectrum underlay, group delay drawn as the sum of the real stages (SPREAD fans the bumps out, STAGES raises them) with a tick per stage centre on the baseline. The vertical viz starts as the oscilloscope (right-click flips to the spectrum). Smoke tests follow the new geometry Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ones pair up only when there are more than three parameters), so the diffusor shows FREQ | Q plus normal STAGES and SPREAD cells; the scope's hover label is two lines (SCOPE / 250ms) so it no longer clips out of the 62 px viewport Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…/ ping-pong, 2 repeat); a tiled drawn curve is baked through eval_tiled (the input is folded or wrapped back into [-1, 1], so mirror is always continuous and repeat seams like Wrap) and flagged periodic so inputs past the table are range-reduced. In the engine TILE only acts on the drawn curve (shape CUSTOM) and forces the saturator to x4 oversampling at least; the oversample dropdown shows the effective rate. Other plug-ins are untouched (tile stays 0, golden tests unchanged). 3 new tests: fold / wrap values and seam continuity, forced oversampling (latency) on the drawn curve only, tile changes what a hot drawn curve does Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ges the DC the shaper adds, and the 5 Hz DC blocker and the 25 Hz asym high-pass then ring on that step (a bump after the 10 ms table crossfade; worst on Crossover / Rectify / Diode / Asym Fold and with bias). The saturator can now estimate each table's DC on the recent input (Hann-weighted, so a loud low tone does not fake a DC) and take the difference off the blocker input and the asym high-pass input in step with the crossfade, releasing it over about a second; the crossfade is 30 ms in SatDiff (ShaperLut.fade_steps, Saturator::set_morph_steps). Off by default in the shared saturator, so SatComp / SatXComp / SatLim render exactly as before (golden tests unchanged). Test: seven shape pairs x bias 0 / 25, the slow error after the crossfade stays under 20 % of the rms and the compensation beats leaving it off by more than 2x. Average measured error over all 364 pairs went from 9.3 % to about 1 % Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ator input. The tile mode now reaches the shaper table for every factory shaper (ShaperLut.build_tiled: the shaper's [-1, 1] stretch is folded / wrapped by tile_x, so a bias sweep walks through copies of it); the periodic shapers (Tri Fold, Sine Fold, Wrap, Asym Fold) already tile themselves, so TILE is ignored there (DiffParams::effective_tile, tile_applies) and costs no oversampling. The forced x4 oversampling follows the effective tile. The input peak now taps the saturator's input (after PHASE and a PRE diffusor, before DRIVE): the dry input while those do nothing, so one meter checks the input level and what reaches the shaper. 3 tests: tile works on a factory shaper and is a no-op on Wrap, forced oversampling exceptions, the meter ignores DRIVE and POST fx but sees a PRE diffusor Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… curve, so off / mirror / repeat draw differently under drive and bias; one point per pixel column while tiled), the three tile buttons dim and go inert on the periodic shapers, the oversample dropdown shows the effective rate. Smoke test: the plot differs between the three tile modes, Wrap ignores the switch Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…r reads after DRIVE. The DC compensation of the previous commit is now continuous instead of planned per shape change: every 16 samples while drive, bias or the shaper table moves, the shaper's DC on the recent input (512 samples, Hann-weighted, before DRIVE / BIAS) is evaluated with the previous and the current settings and the difference is taken off the DC blocker and asym high-pass inputs, so it follows a bias drag, a drive sweep, a shape / tile change and a curve edit with no lag and in silence (where the DC is just the shaper at the bias). Switching the saturator on after a stretch off re-settles the blocker and oversampler (a step from a stale blocker state was the other half of the bump). Dragging BIAS in silence to +-40 %: output peak 0.08 -> 0.0007 (-41 dB to -63 dB). The input meter now peaks on |shaper input| after the HP / LP split, DRIVE and BIAS (0 dBFS = the edge of the drawn curve, so it shows whether the SAT stage is clipped), the plain input while the saturator is off. Still off by default in the shared saturator (other plug-ins unchanged, golden tests pass) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…scene B (shape_b, curve_b, tile_b) next to scene A; with MORPH DEPTH at 0 the scene being edited (ab) is the only one heard (scene A is exactly the plain shaper, as before). With DEPTH above 0 both sound as a table blend (ShaperLut::set_blend, Saturator::set_scene_b / blend_now: one shaper evaluation per sample, the blended table feeds the same DC tracking so a moving blend does not thump). Modes (morph_dir): 2 = MIX, a static crossfade by DEPTH with no detector (the new default); 0 = A>B and 1 = B>A, where a level follower on the saturator's input moves the position (follower.rs: SPEED 100 % = fast peak, 0 % = slow RMS, level -40..-4 dBFS sweeps 0..1, depth scales the excursion; A>B: depth * level, B>A: depth * (1 - level)). DiffEngine::morph_pos exposes the position. 6 tests: the edited scene is heard exactly at depth 0, MIX 100 % = scene B and MIX 50 % = the average of both renders, detector direction and level, SPEED changes the follow time more than 10x, a blend moving through two very different-DC scenes stays tame Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…Tile B (45), scene B's drawn curve (persisted, saved in presets as curve_b), Morph Mode 0..2 defaulting to MIX. The shape selector, tile buttons, ASYM, the draw pencil / editor and the plot all follow the A | B switch (scene A's slots 1 / 38 or scene B's 44 / 45); presets and the modified marker still address both scenes. MORPH block: click cycles MIX (A <-> B arrows, DEPTH cell reads MIX), A > B, B > A; SPEED is greyed and inert in MIX. The fx type dropdown opens from a click or right-click anywhere on the button (not just the arrow). Smoke tests: whole-button dropdown, SPEED inert in MIX and live in the detector modes, edits routed to the scene on the A | B switch Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… heard; segmented meter on the sat output; centre-tapped MIX. DiffParams::sat always uses scene A (the editor's A | B only picks which scene the selector, tile, ASYM, draw editor and plot show; scene B is heard only through MORPH, DEPTH 0 = scene A exactly, MIX 100 % = scene B). The hatched meter now peaks on the saturator output (after SAT MIX and SAT GAIN, before a POST fx) and its clip lamp holds 1 s on its own, no click to reset (the shaper-input tap and xin_abs are gone). MORPH block: the header reads XFADE in MIX mode (MORPH stays for the detector modes, no more repeated word) and the MIX cell is centre-tapped (bar grows from the middle: left = scene A, right = scene B, notch at the centre). Tests: ab never changes the rendered audio, the meter follows SAT GAIN / SAT MIX / clipping and ignores OUT GAIN Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…rst two columns, PHASE is demoted to a small slider row (label left, value right) beside it, and the cell it freed is SAT GAIN, a normal cell: the inverted fill is reserved for OUT GAIN Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… step 4, DSP side). A Curve now holds up to PT_CAP = 30 points (the SatComp-family editors keep their 10-point UI cap, MAX_PTS, and odd_half / even_half still truncate to it) and carries a bend per segment plus a linear flag. A bend lifts the middle of a segment as a parabola that is zero at both points (handle_base gives the handle's spot, the curve passes through it), linear draws straight segments instead of the smooth monotone cubic, the result stays inside [-1, 1]; from_half builds an odd or even curve from a right half with its bends (mirrored: odd -c, even +c, the centre segment of an odd curve stays 0 so it passes through the origin), right_half reads one back, ext / from_parts are the persisted form (empty for a plain curve, so other plug-ins' states are byte-identical). Sanitiser keeps bends with their points. 3 tests: handles, mirrored bends, 30 points Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… even views, counter n/30), curve handles (a hollow circle on the curve in the middle of each segment: drag it vertically to bend the segment, double click to straighten it; in the odd / even views the mirrored half has its ghost handle and an odd curve's centre segment has none), LINEAR | CURVE now acts and is stored per scene (straight segments or the smooth monotone cubic, bends on top of either), inserting a point splits its segment into two straight ones, removing one merges them. Persistence: custom-curve-ext / custom-curve-ext-b (linear flag + one bend per segment, empty for a plain curve, so older projects load unchanged) and curve_ext / curve_ext_b in presets (optional fields, version stays 1). Smoke test: drag a handle, it is persisted, double click straightens, LINEAR writes the flag Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…r reads the dry input again (IN lamp, 1 s hold). MORPH: a thin centre-tapped line along the bottom of the toggle shows which way the blend leans right now (telemetry from DiffEngine::morph_pos; only the centre notch while DEPTH is 0), right click opens the mode list (MIX / A > B / B > A) while left click still cycles, the group is 111 px tall = half the plot and so is the vertical viz (4 px gap). Plot: the other scene as a faint underlay under the edited one (same drive / bias), a 10 x 10 secondary grid of sparse faint dashes in both the normal and the draw view. Meters: taller clip lamps with IN / OUT written inside. Oversample label shows 4x* when TILE forces x4 over a lower selection, so a stuck-looking label is explained Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ut gain; the colour fx do not change dynamics much). IN GAIN (slot 27, id in_gain, -24..+24 dB, smoothed) is first in the chain, before PHASE and a PRE fx; the dry IN meter and bypass / delta read the untouched input. The plot's left-edge marker, wheel and drag now move IN GAIN (label IN +x dB). The saturator's loudness readout / click-to-match compares the saturator output with what entered it after IN GAIN, and a finished match writes OUT GAIN (the cancelling gain). Cell reads IN GAIN (normal colour; the inverted fill stays with OUT GAIN). Test: the gain applies first, meter / bypass ignore it, the match reflects the saturator's own change (not the 6 dB of gain) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…he DIFFUSOR (color.rs, ColorFx behind the fx type switch; Diffusor untouched). Each type is the wet signal, the global COLOR MIX blends it with the dry; one type runs at a time and a type change fades the old one out, clears the new one and fades it in; no latency from any of them (free-running LFO, standard flanger, no through-zero). PHASER: 2..12 first-order allpass stages (the STAGES percentage, 2 per step) swept by a sine LFO around FREQ (DEPTH 100 % = +-4 octaves), FEEDBACK around the cascade (tanh-limited), WIDTH = LFO offset between the channels. FLANGER: modulated delay (centre DELAY 0.1..10 ms, DEPTH as share of it, 4-point cubic read), feedback of either sign, WIDTH. RINGMOD: internal carrier FREQ 20 Hz..20 kHz, SHAPE sine / triangle / square / saw (PolyBLEP on the last two), AMOUNT from the plain signal through AM to ring modulation. New params fx_rate, fx_depth, fx_fb, fx_delay, fx_shape (slots 46..50). 10 tests: bypass exactness for every type, stage counts, phaser flat alone and notching mixed, sweeping, feedback bounded at +-95 % / 20 Hz sweeps / 100 stages, flanger comb notches and the delay time to the sample, ring modulation sidebands without carrier leak, carriers bounded, a type change fades without a click, every type changes the chain and adds no latency Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ur grid: DIFFUSOR FREQ|Q STAGES SPREAD; PHASER FREQ|RATE DEPTH|STAGES FDBK|WIDTH (six parameters, three dual cells); FLANGER DELAY|RATE DEPTH|FDBK WIDTH; RINGMOD FREQ SHAPE AMOUNT. The fx viewport draws the phaser's dry + wet magnitude with its sweep range as a bar, the flanger's comb (with feedback) and the ring modulator's carrier as it modulates. RATE and DELAY taper logarithmically in the editor like FREQ (host ranges stay linear) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…drive, bias, HPF / LPF, tone and ASYM (host params Drive B / Bias B / Sat HP B / Sat LP B / Tone B, slots 51..55, and a persisted per-shaper ASYM list, asym_b; SAT MIX stays shared) next to its shape / curve / tile. Because the scenes now differ before the shaper (different filters, gain, bias), MORPH blends the outputs of two complete saturator chains (scene B's runs only while the crossfade is above 0, is re-settled when it joins, and is crossfaded with a 20 ms-smoothed position) instead of blending shaper tables; the LUT-blend code is removed from the shared saturator. The level-match reference is the blended output; a tile in either scene forces x4 oversampling in both (equal latencies, and the reported latency no longer depends on DEPTH). Presets carry asym_b. 2 tests: MIX 100 % is scene B with its own settings, MORPH off is scene A untouched by B, 50 % is the average of the two chains; a tile in scene B keeps the latency fixed whatever the depth Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ad sections that are not the identity. `fill` pads every chain to MAX_SEC (8) sections, and both ran all of them every sample: ~120 + ~128 biquads per sample at the default 24 dB slope, of which ~15 + ~12 did anything. Each chain now keeps `act`, the length up to the last section that is not the identity in the current OR the target coefficients (a glide moves a section between the two), refreshed in set / begin_block; sections that drop out are zeroed so they start clean when they return. xbench (amen18, 512): compressing 214 -> 117 us/block (1.85 -> 1.00 % of a core), drive 30 x1 307 -> 211. The audio is the same to the DENORM constant (max abs diff 8e-29) in 10 of 12 xbench variants, 2e-10 with 96 dB slopes, and 3e-5 peak (-70 dB re the signal) for ~2 blocks after an automated slope change, where the old code ran leftover filter state through the identity padding Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…off and the filter has rung out. With DRIVE and BIAS at 0 the saturator returns the clean signal for both outputs, so the "added content" fed to the band-limiter is exactly zero and the 2-4 biquads per channel per band filtered silence. Strip::begin_block sets `confine_idle` = saturator off AND every filter state below 1e-12 (so a drive that was just turned to 0 still gets its tail), and snaps the idle filter's coefficients to their targets (nothing advances the glide while it is skipped, so a later switch-on starts with the right filter). Saturator::is_on is the only change outside satxcomp-dsp. xbench compressing 117 -> 101 us/block (1.00 -> 0.87 % of a core; 214 -> 101 with the previous commit, -53 %). Same audio as the old code to 1e-28 in every variant, including a new one that switches bands' drive / bias on and off while the music plays; workspace tests green Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…cing) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ot matter, bit-exactly. The gain computer returns exactly 0 dB at or below the foot of the knee, so a detector level under `lv_floor` (the linear equivalent, set in derive) goes straight to 0 without the `ln`; and exp(0) is exactly 1, so a channel with no gain change (g == lift) skips the `exp`. Both are the common case for a compressor that is not working. SatXComp xbench (4 bands): defaults 94 -> 71 us/block, compressing 100 -> 92, hard compression (-45 dB, 10:1) 121 -> 116; the output is identical to the bit in all 14 xbench variants Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…padding, idle Confine, compressor short-circuits), the dropped fast-math experiment, and how to profile and regression-check with xbench Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… for the shared core (satcomp-dsp): median us per 512-sample block of one stereo Saturator for x1 / x2 / x4 / x8, with and without the even-part (asym) path, odd vs asymmetric shapers, HP / LP / tone / mix, and SatDiff's DC compensation with the drive held still and moving. XB_DUMP=dir writes each variant's output and `satbench cmp dirA dirB` compares two dumps, like xbench Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…pers morphing) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…, and SatLim's) are polyphase with linear history buffers. The old ones summed 24 taps in one serial chain through masked ring indices; the taps are symmetric, so Hb is now const-generic over its length, keeps each history twice (written at p and p + K, so the last K samples are one contiguous slice), folds the symmetric taps pairwise (half the multiplies) and sums over four independent accumulators. `Taps` carries the first half of each filter. The old stages stay in the tests as the reference: the new ones match to < 1e-12 for both lengths (K = 24 and 8), up and down. satbench, one stereo saturator, 512 samples: x2 50 -> 22 us, x4 58 -> 37, x8 88 -> 68; output differences <= 2e-9 (below the f32 resolution of the output); workspace tests green Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… odd shapers. Every saturator splits the shaper into an odd and an even part (ASYM) and runs the even part through its own set of down-filters plus a second LUT evaluation: nearly double the work of the legacy path. Seven of the 15 factory shapers (Soft, Rational, Atan, Hard, Tri Fold, Sine Fold, Crossover; also a symmetric custom curve) are exactly odd: their even part is <= 1e-16. ShaperLut caches `odd` (|even part| < 1e-12 over the whole table; set when a table is committed or a morph ends, false while morphing); Saturator::begin_block sets `even_skip` when the table is odd AND both channels' even-part down-filter histories have rung out (< 1e-12), so the tail of a previous asymmetric shaper still plays out exactly as before; while it is set tick_split uses the plain `process` and e = 0 (the 25 Hz even high-pass and the DC compensation offsets still run). satbench, one saturator, drive 30 %: Soft x1 11.5 -> 9.2 us, Soft x4 37.8 -> 21.2; Tube unchanged. Output identical to the bit in every steady variant; the shape-switching ones (odd <-> tube, morphing, with and without dc_comp) differ in 3-4 samples of 600 000 by one f32 ulp; workspace tests green Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… every 2nd history sample, every 32 samples instead of every 16. While DRIVE, BIAS or the shape moves it re-evaluated all 512 history samples through two tables per channel every 16 samples (~250 table lookups per output sample): 13 us/block with the drive still, 199 us (1.7 % of a core) while it moved, which is what parameter automation or a slider drag costs. Only the difference between the old and the new parameters is used, taken over the same points, so thinning the points mostly cancels; the look interval is a multiple of the stride so the points are the same at every look. Measured against a reference tracker (full history, every sample) on the amen loop with the drive moving: sub-20 Hz error 3.8e-3 peak (-41.5 dB re the signal; -48 dB before), far below the thump the compensation removes; SatDiff's shape-switch thump test moves by <= 0.00002 (worst 0.066, limit 0.2, compensation still beats none by the same margin). Stride 8 / every 64 was tried and dropped (error up 25 % in that test, -30 dB on music). satbench: drive moving x1 199 -> 65 us, x4 253 -> 92; drive still unchanged. Workspace tests green Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…haper skip, DC tracker thinning), with the lesson that the tracker has to be judged on music and not on a sine Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… label contains it. A single case wobbles by 10-13 % inside the full suite (a SatLim "+13 %" turned out to be 1.09-1.18 % against 1.11 % when repeated four times on its own), so a suspected regression has to be repeated before it is believed Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…Lim, hover on every switch, new header lockup Draw view: MAX_PTS 10 -> 20 (10 per half in the odd / even views; the Curve already holds 30, SatDiff keeps its own 30 and the bends / LINEAR / TILE). A third tool next to the pencil and the odd tool mirrors the drawn half across the Y axis, f(-x) = f(x): the ghost half, adding, dragging and picking follow the mirror's sign; switching between odd and even keeps the drawn half, free -> odd / even keeps that part of the curve. Tools sit pencil, odd, even left to right (as in SatDiff) and the title reads DRAW FREE / ODD / EVEN. even_glyph is shared in satcomp-ui. Presets with more than 10 points still load in older builds (the sanitiser keeps 30; only adding is capped there). Tests: SatLim gets the even-tool and the 20-point tests; SatXComp's draw-view test follows the new tool positions and covers the even tool; both now run on the Custom shaper (the test host does not apply the view's own write, which closed the view after a few frames and made the closing assertion vacuous). Hover: Widgets::btn fills the button under the pointer (a lit one gets an inner outline, as SatDiff's hbtn), over_btn colours hand-drawn icons over an empty-labelled button, the click-to-cycle SRC / WAVE cells under every scope fill on hover (SatDiff too), SatXComp's band buttons and EQ toggle and SatComp's EQ tabs light their squares and labels under the pointer. Header: the lockup follows plugintitles/*.svg at 150 %. A 20 x 20 plate (10, 6) with a per-plug-in icon (Canvas::logo_art: SatComp / SatXComp arrows, SatLim brackets, SatDiff diagonals), the name in a 18.5 title font with the middle of its capitals on the icon's centre, the version in a 10 px font at half opacity with its digits on the title's baseline. Both are measured from the font (Canvas::ink_span reads the glyph's ink rect), so every face lines up whatever its ascent; the version is left out when it would run into the neighbouring button (widest fonts, SatXComp's save button). The hover plate covers the whole lockup. SatComp's UI golden is regenerated (every frame: the header changed). Workspace tests green Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…le button on the column edge
Factory presets: Amen Hell, Bump Me Up, Burnt Popcorn, Shrooms, Soft2Hard, Tube2Brick, You Rang and Zap, taken from the owner's user presets (values rounded, curves, handles and ASYM kept; file names use underscores like SatLim's). Test: all eight are in the bank, every value is inside its parameter's range, the curves / handles / ASYM list ride along.
About page: the alpha copy ("phase rotator and diffusor are coming") is replaced with the tone of the other three plug-ins and describes the drawable curve (30 points, handles, LINEAR / CURVE, odd / even, TILE), the A | B scenes with MORPH, PHASE and the colour slot (PRE-SAT / POST-SAT, COLOR MIX), LIMIT and GAIN MATCH; the v0.1 changelog entry follows.
The window-scale button moves 2 units left so its right edge (236) lines up with the preset arrows and the shape selector below it; the title lockup's width limit follows.
Workspace tests green
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…m 0.1.1 (SatDiff debuts at 0.1.0) Cargo versions of the plug-ins and their dsp / ui crates (the plug-ins' header, About and the embedded manual take the version from Cargo, so the manual PDF has to carry the new version in its name: SatComp-Manual-v0.5.3.pdf, SatXComp-Manual-v0.1.1.pdf, SatLim-Manual-v0.1.1.pdf are copies of the previous PDFs for now, rebuild them with docs/manual/build.py after the manual text is updated). The manual sources, README and INVENTORY carry the new versions. In-app changelog entries for the 20-point draw view with the EVEN tool, hover highlights, the new header, and the faster saturator filters; SatComp's UI golden is regenerated for the extra changelog entry. Cargo.lock follows. Workspace tests green Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…was left on AppState keeps `about_last` (About, Changelog or Preferences) for the editor session: the title toggle (AppState::toggle_about) closes the page remembering the tab, and opens it on that tab again (About the first time); the X CLOSE button records it too, and reopening on Preferences scrolls the scheme list to the current scheme as the tab click does. Not persisted across sessions. SatLim smoke test: Preferences, close, reopen = Preferences; Changelog likewise. Workspace tests green Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… 0 dB lines, dB legends and a draggable curve The plot's range is now a persisted view setting (ceiling 0..+18 dB, floor -12..-72 dB, default +6 / -48), so a level past 0 dBFS is visible; dashed lines mark 0 dB on both axes and a dashed line drops from the threshold handle. CEIL / FLOOR labels in the lower right drag, wheel and double-click-reset. The line above the threshold drags the ratio and the line below drags the threshold; the ratio handle no longer sits under the scope icon at 1:1. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… SatDiff manual, refreshed PDFs build.py expands <!--flow:name--> and <!--hero:name--> from figures.py / figure_specs.py: the flowcharts are rects and arrows like the live About flow, the hero is the screenshot (shadow cropped) with leader lines, module brackets and notes. SatComp's chain line was wrong (the engine runs SAT, PRE EQ, COMP, MIX, POST EQ, OUT GAIN, LIM) and is fixed. Draw view is 20 points with ODD / EVEN; SatComp documents the plot's CEIL / FLOOR and the draggable curve. SatDiff 0.1.0 gets its first manual. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…adow), SatDiff in the README, old screenshots removed Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…elog mentions the compressor plot; SatComp golden re-recorded (about / changelog frames only) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… suite zip per platform and a macOS .pkg (docs/install/make-suite.sh, run by a new `suite` job on v* tags, attached by the release job); suite INSTALL texts and a Linux install.sh that handles every plug-in; install texts say the themes file is shared by all the plug-ins Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ownloads); INVENTORY gets SatDiff, the new versions and the suite files; README install text Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…computed draw commands differed in 2 frames on Linux), and the test job uses --no-fail-fast so one failing crate does not hide the others Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… .pkg, one joint manual PDF) make-suite.sh joins the four manuals with pdfunite into SatComp-Suite-Manual-<tag>.pdf (in each suite zip and on its own); the release job drops the per-plug-in zips and PDFs when suite files exist (a <plugin>-v* tag keeps its own zip); install texts, README and v0.5.3 notes updated. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…all, appended so saved indexes keep their meaning Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…stroked lines (no protruding diagonals, scales with the window); generated by tools/gen_logo.py; SatComp golden re-recorded; release notes Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…o the four corners (its SVG is asymmetric, which showed as a wonky top-right bracket at 100 %) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…t); README intro reworded Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…rsion>-dev) so the suite zips, .pkg and joint manual can be checked before tagging Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
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.
Release branch for v0.5.3. Don't merge until the Windows and Linux checks pass.
Includes: SatDiff 0.1.0 (new), SatComp compressor plot (CEIL/FLOOR, dashed 0 dB, draggable curve),
20-point draw view + EVEN tool, hover on switches, CPU work, Linux fixes, manuals for all four,
new screenshots, README, suite zip + macOS .pkg (tags only), release notes docs/releases/v0.5.3.md.
To check before merging: