Move AUDIODSP_PIN to audiodsp f98eb17, the 128-voice sampler

AUDIODSP_PIN

@@ -414,5 +414,23 @@
 # promotions in the SVF handover (#264). audiosampler is a new module and
 # touches no node that existed.
 #
+# Moved to f98eb17 on 2026-10-20, for audiosampler's 128-voice ceiling
+# (audiodsp#289), so an instrument can ask the sampler for more than 64
+# voices. Named by commit; no release carries it yet. What else moved
+# beneath the components, by `git log 6db15e3..f98eb17 -- src`: audiosampler
+# keeps its samples in a SampleBuffer outside the collected heap and fades
+# a stopped or stolen voice over 5 ms; audiomixer passes a unity-level
+# voice through unscaled, and on CPython scales any other in C (#273);
+# synthio returns silence for a block where nothing sounds (#278);
+# audioladder takes buffer_size; audiooscillator is new, and audiopump and
+# audiometer change only their USB paths. audioecho.FeedbackDelay's loop
+# shifter now searches a whole window for a splice that nothing near fits
+# (audiodsp#284), but not on a fading repeat (audiodsp#290): #284 alone
+# dropped PitchShifter's third repeat 19 dB at 22.05 kHz, -5 semitones and
+# Feedback 0.3 (T3), and with #290 it reads 2.5 dB over its bar's
+# reference, as it did before #284. The two `@options` rows in
+# test_measure_effect_cost run the loop shifter on held material and move
+# with #284; their digests are re-read there.
+#
 # <ref> <commit>   -- ref is what CI installs (git+https://...@<ref>)
-6db15e311a17ca83784f0276d326f98b77950b5f 6db15e311a17ca83784f0276d326f98b77950b5f
+f98eb17695c1c1e64284a4ef1f2abee7f3cdeefa f98eb17695c1c1e64284a4ef1f2abee7f3cdeefa

CHANGELOG.md

@@ -48,6 +48,12 @@ there, and are recorded in its changelog.
 
 ### Changed
 
+- The gates run against audiodsp f98eb17, where the sampler's voice ceiling
+  is 128, so a SoundFont pool built with `max_voice_count` above 64 gets the
+  voices it asks for. The loop shifter that `PitchShifter`'s feedback runs
+  through now searches further for a splice that fits a held chord
+  (audiodsp#246), but not on a fading repeat, so the repeats keep their
+  level (audiodsp#290).
 - A SoundFont `Voices` pool makes its sampler with the voices it was asked
   for, not the build's ceiling. Since audiodsp raised that ceiling to 128, a
   48-voice pool held about 32 KB more heap on MicroPython (68 KB, from 36 KB)

tests/test_measure_effect_cost.py

@@ -42,10 +42,13 @@ def test_the_damping_row_turns_the_loop_low_pass_on(self):
         # as a chord each change its bytes, and both together give these.
         # Before v0.6.6 it was e7a07359012d7a8e, and 9e06dd6c5d8a2f38 with
         # damping. The base and fb0.9 rows run no shifter and did not move.
+        # audiodsp#284, a splice that nothing near fits searching on out to
+        # a whole window, moved both again; they were 7dfa014ed463ea16 and
+        # 101319c53132701d before it.
         options = digest("audioecho.FeedbackDelay@options")
-        self.assertEqual(options, "7dfa014ed463ea16")
+        self.assertEqual(options, "845477cccfdef016")
         self.assertEqual(digest("audioecho.FeedbackDelay@options+damping"),
-                         "101319c53132701d")
+                         "181eec0ad19eae28")
 
 
 class TheBesideRow(unittest.TestCase):