New reference to show off a realsense viewer. by nicksinas · Pull Request #20 · avocado-linux/references

nicksinas · 2026-07-15T03:30:11Z

No description provided.

jetm

Compared against the existing python-flask/python-yolo reference pattern; one correctness concern on the load-bearing dependency, plus a minor doc-drift note.

…ewer

Copilot

Pull request overview

Adds a new realsense reference project that demonstrates a native C++ Intel RealSense “web visualizer” system extension for Avocado OS, including build tooling, runtime packaging, and end-user docs.

Changes:

Introduces a new realsense/ reference with README + getting-started documentation.
Adds a C++17 RealSense capture + libmicrohttpd-based MJPEG/JSON server with an embedded dashboard UI.
Adds Avocado extension configuration, systemd service unit, and compile/install/clean scripts.

Reviewed changes

Copilot reviewed 10 out of 12 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
realsense/README.md	Reference overview and metadata (targets/topics/icon) for the references index.
realsense/getting_started.md	Step-by-step instructions for init/build/deploy/verify/customize.
realsense/avocado.yaml	Defines supported targets, extensions, packages, and SDK compile steps.
realsense/avocado.lock	Locks distro/targets for the reference (currently only raspberrypi5).
realsense/app/src/CMakeLists.txt	CMake build for the `realsense-visualizer` executable linking to packaged deps.
realsense/app/src/app.cpp	Implements capture loop, HTTP endpoints, MJPEG streaming, and dashboard HTML.
realsense/app/overlay/usr/lib/systemd/system/realsense-visualizer.service	Installs/enables the systemd service to run the visualizer on boot.
realsense/app-install.sh	Installs the built binary into the extension sysroot via `cmake --install`.
realsense/app-compile.sh	Cross-compiles via generated toolchain file and CMake build directory.
realsense/app-clean.sh	Cleans build artifacts.
realsense/.gitignore	Ignores Avocado state and build output for this reference.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

jetm

Traced all of Copilot's findings against the full app.cpp - they're all valid and worth fixing: feed_reader returning 0 (L256) is non-idiomatic and version-fragile for the MJPEG path; MHD_create_response_from_buffer/_from_callback can return NULL (L280/L294) and are used without a guard (the callback path also leaks the FeedState on that NULL); and /api/distance (L561) returns 503 for missing x/y query params, which is a client error that should be 400, not "no depth data." The README targets vs avocado.yaml supported_targets mismatch is real too - the manifest adds imx8mp-var-dart, imx95-frdm, and ucm-imx8m-plus that the README omits. The one thing not already flagged is the most serious (inline at L175): the capture-thread loop body has no exception handler, so a camera unplug mid-stream aborts the whole process. Lower-priority: g_running is never cleared (no signal handler, so the shutdown atomic is inert), libjpeg's default error_exit calls exit() on a fatal, and app-install.sh writes to DESTDIR=$AVOCADO_BUILD_EXT_SYSROOT without guarding the var is set.

- capture loop: wrap start + stream body in try/catch(rs2::error) so a mid-stream camera fault (USB unplug) reconnects instead of terminating - feed_reader: never return 0 mid-stream (MHD reads that as end-of-stream); loop until a frame is ready or shutdown - guard NULL from MHD_create_response_from_buffer/_from_callback; free FeedState when MHD doesn't take ownership - /api/distance: 400 for missing x/y params, 503 only when depth is absent - clamp the distance lookup against the depth buffer dims, not intrinsics - SIGTERM/SIGINT handlers so shutdown actually clears g_running - libjpeg: custom error_exit via setjmp so a bad frame drops the JPEG instead of calling exit() - app-install.sh: fail clearly if AVOCADO_BUILD_EXT_SYSROOT is unset - README: match frontmatter targets to avocado.yaml supported_targets

nicksinas · 2026-07-18T15:38:55Z

Pushed bc693a1 addressing the review. All inline threads (Copilot's five + the capture-loop crash and latent OOB) are fixed and resolved.

The three lower-priority items from the review summary are also in this commit:

g_running was inert — added SIGTERM/SIGINT handlers so a systemd stop clears the flag and the capture loop exits cleanly.
libjpeg error_exit calling exit() — installed a custom error manager (setjmp/longjmp); a bad frame now drops the JPEG and the feed continues instead of the process exiting.
app-install.sh unguarded DESTDIR — it now fails clearly if AVOCADO_BUILD_EXT_SYSROOT is unset.

Note: I've verified every supported_targets entry (including imx8mp-var-dart, imx95-frdm, ucm-imx8m-plus) has librealsense2 + -dev in the feed, and the README frontmatter now matches the manifest. Still pending a real avocado build on an aarch64 target to confirm the cross-compile before merge.

Copilot

Pull request overview

Copilot reviewed 10 out of 12 changed files in this pull request and generated 4 comments.

Comments suppressed due to low confidence (1)

realsense/app/src/app.cpp:350

/api/distance uses std::atoi, which silently treats non-numeric inputs as 0 (e.g. x=foo => x=0). That makes invalid client requests look like valid measurements. Parse with strtol and return 400 on conversion failure.

    int x = std::atoi(xs);
    int y = std::atoi(ys);

- json_escape: escape control chars (\n, \r, \t, \b, \f, <0x20) so device info strings can't produce invalid JSON - dashboard: build deviceBar with textContent instead of innerHTML, closing the XSS surface from /api/device values - feed_reader: encode only new frames via a frame sequence counter rather than a fixed 33ms sleep, matching capture rate (no duplicate encodes) - /api/distance: parse x/y with strtol, return 400 on non-numeric input instead of silently treating it as 0 - avocado.yaml: label the empty root password NOT FOR PRODUCTION

nicksinas · 2026-07-18T15:57:17Z

Pushed e148048 for the second review pass — all four inline threads resolved (json_escape control chars, deviceBar XSS via textContent, duplicate-encode throttle, NOT FOR PRODUCTION password note).

Also addressed the low-confidence suppressed item: /api/distance now parses x/y with strtol and returns 400 on non-numeric input instead of letting atoi silently coerce it to 0.

Copilot

Pull request overview

Copilot reviewed 10 out of 12 changed files in this pull request and generated 3 comments.

- jpeg_encode: establish setjmp before jpeg_create_compress so a longjmp can never land on an uninitialized buffer; guard destroy with a created flag and mark it volatile across setjmp - /api/distance: reject out-of-int-range x/y with 400 before narrowing the strtol result (add <limits>)

Copilot

Pull request overview

Copilot reviewed 10 out of 12 changed files in this pull request and generated 3 comments.

- serve the static dashboard HTML as a persistent MHD buffer instead of copying the whole page on every "/" request - build g_device_json with std::string instead of a fixed 512-byte buffer so a long device string can't truncate it into invalid JSON - getting_started.md: note the empty root SSH password is dev-only, not for production images

jetm

This revision addresses almost everything from the last round - NULL/lifetime checks on the MHD responses, the content-reader no longer returns a bare 0, 400-vs-503 is cleanly separated, the clamp/index dims are consistent, g_running is now driven by a real SIGTERM/SIGINT handler, the README targets match avocado.yaml, and the device bar switched to textContent (nice XSS close). Two residual items inline. One more (LOW, no clean line to anchor): app-install.sh got the AVOCADO_BUILD_EXT_SYSROOT guard this round, but app-compile.sh still expands OECORE_TARGET_SYSROOT/CROSS_COMPILE unquoted with no guard (set -e, not -u), so run outside the SDK env it emits a silently-wrong host toolchain file instead of failing - add a matching -z check for parity.

- capture_loop: also catch std::exception (e.g. std::bad_alloc from the ~1MB per-frame buffers under memory pressure) so it can't escape the thread and call std::terminate; factor the stale-mark + backoff into a shared recover() used by both handlers - jpeg_encode: qualify out/out_size volatile so their values stay defined across setjmp/longjmp (C11 7.13.2.1) before the recovery-path free

jetm

The two inline fixes from last round are verified good - the capture thread now catches std::bad_alloc, and out/out_size are volatile with the const_cast at jpeg_mem_dest. Three items inline: the same bad_alloc hazard is unguarded on the HTTP-callback side (feed_reader runs in an MHD worker thread) - MEDIUM; app-compile.sh still lacks the env guard app-install.sh got; and the capture-thread config setup runs just outside its try. Everything else re-verified clean (MHD NULL/lifetime, 400-vs-503, thread-safety/bounds, signal-driven shutdown, README/yaml match).

- feed_reader / handle_request: wrap bodies in try/catch(...) so a std::bad_alloc (from snapshot copy, jpeg_encode, or chunk concat) can't unwind into libmicrohttpd's C worker thread; end the feed with MHD_CONTENT_READER_END_WITH_ERROR and return MHD_NO for the request - capture_loop: construct rs2::config inside the try so a throw during stream config can't escape the thread either - app-compile.sh: guard OECORE_TARGET_SYSROOT (mirrors app-install.sh) and quote the sysroot path; fail early with a clear message outside the SDK env

jetm

Reviewed across several rounds; all findings are addressed and verified on 6e3a86c - the HTTP-callback threads (feed_reader/handle_request) now catch exceptions, app-compile.sh guards OECORE_TARGET_SYSROOT, and the capture config setup is inside the try. LGTM.

New reference to show off a realsense viewer.

916087c

nicksinas requested a review from jetm July 15, 2026 03:30

jetm reviewed Jul 15, 2026

View reviewed changes

Comment thread realsense/app-compile.sh Outdated

updates to the reference to move to cpp to show off only realsense vi…

59e23a7

…ewer

nicksinas requested review from Copilot and jetm July 17, 2026 21:35

Copilot started reviewing on behalf of nicksinas July 17, 2026 21:35 View session

Copilot AI reviewed Jul 17, 2026

View reviewed changes

Comment thread realsense/app/src/app.cpp Outdated

Comment thread realsense/app/src/app.cpp

Comment thread realsense/app/src/app.cpp

Comment thread realsense/app/src/app.cpp Outdated

Comment thread realsense/README.md

jetm reviewed Jul 17, 2026

View reviewed changes

Comment thread realsense/app/src/app.cpp Outdated

Comment thread realsense/app/src/app.cpp Outdated

nicksinas requested review from Copilot and jetm July 18, 2026 15:40

Copilot started reviewing on behalf of nicksinas July 18, 2026 15:41 View session

Copilot AI reviewed Jul 18, 2026

View reviewed changes

Comment thread realsense/app/src/app.cpp

Comment thread realsense/app/src/app.cpp Outdated

Comment thread realsense/app/src/app.cpp Outdated

Comment thread realsense/avocado.yaml

nicksinas requested a review from Copilot July 18, 2026 16:03

Copilot started reviewing on behalf of nicksinas July 18, 2026 16:04 View session

Copilot AI reviewed Jul 18, 2026

View reviewed changes

Comment thread realsense/app/src/app.cpp Outdated

Comment thread realsense/app/src/app.cpp

Comment thread realsense/app/src/app.cpp

nicksinas requested a review from Copilot July 18, 2026 17:36

Copilot started reviewing on behalf of nicksinas July 18, 2026 17:37 View session

Copilot AI reviewed Jul 18, 2026

View reviewed changes

Comment thread realsense/app/src/app.cpp Outdated

Comment thread realsense/getting_started.md

Comment thread realsense/app/src/app.cpp Outdated

jetm reviewed Jul 20, 2026

View reviewed changes

Comment thread realsense/app/src/app.cpp Outdated

Comment thread realsense/app/src/app.cpp

jetm reviewed Jul 20, 2026

View reviewed changes

Comment thread realsense/app/src/app.cpp

Comment thread realsense/app-compile.sh Outdated

Comment thread realsense/app/src/app.cpp Outdated

jetm approved these changes Jul 20, 2026

View reviewed changes

Uh oh!

Conversation

nicksinas commented Jul 15, 2026

Uh oh!

jetm left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jetm left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

nicksinas commented Jul 18, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

nicksinas commented Jul 18, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jetm left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

jetm left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jetm left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants