Repository navigation
Conversation
… non-RT retirement
…nd concurrent retirement
… active pins and epoch ACK
…of falling through to CH1
…allback without changing legacy CH1 default
…ng active dry multi-input FAT
…th ACK and diagnostics
…e on return to CH1
…epoch safety CTest
…fore-apply guidance
…poch and original CH1 VST behavior
…ical FAT acceptance
…out regressing native progress
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a311c36ab2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if(readers_.load(std::memory_order_acquire)!=0) return false; | ||
| for(std::size_t i=0;i<retiredCount_;++i) retired_[i].reset(); |
There was a problem hiding this comment.
Synchronize graph reclamation with concurrent reader entry
During concurrent publication and audio processing, this acquire-only reader-count check does not establish sufficient ordering with active_ publication and ReadLease entry: the control thread can observe zero readers while a reader still acquires a retired snapshot, then free that snapshot underneath it. I reproduced a heap-use-after-free in SpyGraph::value by running the new publisher test with AddressSanitizer and increasing its concurrent publication loop from 1,000 to 1,000,000 iterations. Use a proven reclamation protocol with the required cross-atomic ordering so callback-held snapshots cannot be destroyed.
Useful? React with 👍 / 👎.
| api.emitEvent("arMixAction",{action:"routingDryApply",value:{ | ||
| deviceEpoch:Number(d.epoch)||0,revision:Number(d.revision)||0, | ||
| channels:draftRoutes.map(c=>({...c})) |
There was a problem hiding this comment.
Bind routing drafts to the device epoch they were drawn from
If the device changes while the routing inspector is open, arMixStatus replaces state and updates the displayed epoch, but the dropdowns and draftRoutes still describe the previous device. APPLY then sends those old assignments with the new device's epoch, bypassing the native stale-device rejection. When both devices have the same numeric port IDs, the command succeeds against different physical inputs without the operator remapping them. Capture the epoch when drawing the draft and invalidate or rebuild the assignments when that epoch changes instead of substituting the latest epoch at submission.
Useful? React with 👍 / 👎.
…nd reject stale publish
…/writer gate, prevent UAF
…ny stale port intent
|
Cross-thread reconciliation 2026-10-28: do not merge this stale branch wholesale. It has Windows CI PASS but UI smoke FAILURE 40608546037, and predates merged #120 fader baseline. The exact P1E-only changed blobs were moved to PR #121, based on current |
|
Superseded and fully integrated via PR #121, merged at |
P1E MAJOR MILESTONE — native real hardware 10-channel DRY routing FAT, not another offline prototype
This PR connects the already merged P1D immutable 10-strip physical mono/stereo mixer to the real JUCE audio callback behind an explicit operator-controlled DRY ROUTING LAB. The original CH1 VST3 production path is completely unchanged and remains default whenever LAB is OFF.
Delivered
docs/P1E_NATIVE_DRY_LAB_FAT_20261028.md.Cross-thread reconciliation: starts from merged P1D #114
d30cf35and merges current main4951f3bpreserving P3 Backup #116 and pre-fader/DAG correction #118. Old GUI/FX #102 remains isolated. P0 zero-VST crackle #15 is still a physical release gate.Scope guards
This is an actual live dry-routing path but an opt-in hardware diagnostic, NOT full performance mode. No independent 10× VST3 chains, AUX sends, per-strip production fader binding, automatic external port restore, guaranteed click-free physical operation or hardware-verified output. No final artifact/HW FAT claim without exact-head Windows CTest, Chromium and user physical interface exercise.
Acceptance gate
Refs #109 #95 #15 #85.