[Nir-Az]

Overview: In the viewer, setting a post-processing filter control (e.g. decimation magnitude, rotation) applied the effect but the control snapped back to its previous value; it now stays at the value the user set.

Tracked on [RSDEV-12488], [RSDEV-12502]

Why

Regression from PR #15029 (viewer-ui-exp-change-avoid-stream-pause), which routed all viewer option writes through the async dispatcher. That async path exists to avoid blocking the UI thread on the FW round-trip — but it never refreshes the control's cached value, relying on the sensor's on_options_changed echo to do so. Software post-processing filters run in-process and emit no such echo, so after the ~2 s user-request mask expired the control reverted to its stale cached value (decimation magnitude → 2, rotation → 0) even though the effect was applied.

Summary

• Software post-processing filter options (processing_block_model) now write synchronously via set_option, which reads the value back so the control reflects what was applied. FW/sensor options keep the async path.
• These filters have no FW round-trip, so a synchronous write can't freeze the UI; an integer-stepped slider only fires a handful of writes per drag. No throttle/timer needed (the pre-async throttle was an FW-round-trip mitigation).
• Added a write_value() helper in option_model that routes by a write_synchronously flag; the flag is set for processing-block options.

Test plan

• New viewer UI test controls/post_processing_value_persists: streams depth, enables post-processing + decimation, sets magnitude to a non-default value, and asserts the cached control value tracks the applied value. Verified on D455 — passes with the fix, fails when the option is forced back onto the async path.
• Added post-processing panel navigation helpers to the viewer test harness.
• LibCI run.

---

🤖 Generated by AI

[remibettan]

Few comments

[sysrsbuild-gh-agentic]

Reviewed the synchronize commit (). The three previously unresolved review threads (stopwatch reset, gated invalidate, hardened helper) are all addressed by this push — good. Two new observations below.\n\n🤖 Auto-generated comment by rs-agentic bot

[sysrsbuild-gh-agentic]

rs-agentic bot — synchronize review (commit a42a490)

change approved — previous bot comment threads

The commit a42a490 ("viewer: scope invalidate gate to a local error buffer, fix stale test comment") fully addresses both outstanding bot comment threads:

1. Stale error_message can suppress invalidation on a successful write

The write_value implementation now uses a local write_error buffer — a stale error from a prior option's draw in the same frame can no longer suppress the invalidate on a successful write.
change approved ✅

2. Misleading test comment — "async wait"

The wait_until comment in test-post-processing-revert.cpp now correctly reads:
"Software-filter writes are synchronous, so the value is already applied by the time set_post_processing_value returns; this poll is just a defensive safety net."
No misleading suggestion of an async wait. change approved ✅

---

New-commit review — no new issues found

The four commits in this PR are clean:

• option-model.cpp — write_value properly resets dev->last_user_set_stopwatch, uses the local error buffer, and gates *invalidate_flag on write success only. Logic is consistent with the set_option_async contract.
• processing-block-model.cpp — get_option_model uses find+iterator correctly (single lookup). write_synchronously = true is set before map insertion.
• viewer-test-helpers.cpp/h — Refactoring of set/get_value_by_seed is clean. enable_post_processing_filter correctly calls enable_post_processing first so is_enabled() reflects the master+per-filter combined state.
• test-post-processing-revert.cpp — Test picks a target that differs from the persisted value, uses wait_until defensively, and handles no-device case gracefully.

🤖 Auto-generated comment by rs-agentic bot

[remibettan]

LGTM