Research window: 2026-10-27 08:00 ~ 2026-10-28 08:00 (Beijing time, 24 hours)

Sources: RT-Thread official news API and community article API, GitHub (main repository commits / PR / Issue / releases / tags, RT-Thread/packages, vscode-rt-smart, sdk, pkgs-test, topic:rt-thread ecosystem search), Google News aggregation (seven queries in Chinese and English, via proxy), HN Algolia, industry media (EET China, Kuai Technology, 21ic) and vendor channel verification (see the source list in the appendix for details)

Index of This Issue

  • I. Kernel and Version Updates
    • After two quiet days, the mainline resumes merging, with 4 commits landed within six minutes (10-27)
    • The UTIME_OMIT check in dfs is inverted, so timestamps that should not be changed are overwritten (10-27)
    • RAW socket protocol validation switches to a whitelist, so protocols such as ICMP are no longer blocked (10-27)
    • poll now obtains fops from dfs_file, compatible with both DFSv1 and DFSv2 (10-27)
    • 8250 earlycon starts reading reg-io-width, so wide-register platforms no longer access by byte (10-27)
    • The release side remains at v5.3.0; the mainline was bumped to v5.4.0 on 10-17 but has not been tagged (10-27)
  • II. Pending Merge Queue and Security Follow-up
    • The signalfd close path adds slot bounds checking and listener unbinding, directly addressing the out-of-bounds issue recorded yesterday (10-27)
    • A whole-kernel clang-format 22.1.8 formatting proposal is under review, involving 75 files (10-27)
    • The ESP32-S3-DevKitM-1 proposal is refreshed at the end of the window, with a cumulative 295371 lines added (10-28)
    • Four architecture proposals for RT-Thread Smart remain in the queue, all submitted on 10-18 (10-27)
  • III. Toolchain, Official and Community
    • A long community article breaks down Env v2.0.3: commands consolidated into rt-env, with a new plugin system and WebUI (10-27)
    • The official news API has no new items in 24 hours; the latest issue remains dated June 17 (10-27)
    • EET China’s comparison of three major RTOSes lists RT-Thread as a preferred choice for low-power and consumer scenarios (10-27)
  • IV. Ecosystem and Vendor Adaptation
    • Arm China WTCC 2026 seminar: RT-Thread focuses on embodied intelligence projects and breaks down three types of Agent solutions (10-27)
    • A kitchen flame detection prototype on K230 releases its source code and reproduction record, built on RT-Thread Smart (10-27)
    • Handwritten symbol recognition using RT-Thread plus LVGL and TFLM on the CH32H417 dual-core RISC-V (10-27)
  • V. Industry Applications and Embodied Intelligence
    • AIoT competition entries turn toward the industrial side: AI monitoring of BLDC motor defects on Titan Board (10-27)
    • Lead for this issue: chip vendors, board vendors and the operating system side appear together under the alliance framework, all pointing to embodied intelligence (10-27)
  • VI. Summary and Trend Observations
    • Trend for this issue: merging restarts, with fixes concentrated on boundary validation rather than new features
    • Limitations and deduplication notes
  • Appendix: Source verification table and source list

I. Kernel and Version Dynamics

1.1 Mainline resumes merging after two quiet days, 4 commits land within six minutes (10-27)

Date: 2026-10-27 19:20 ~ 19:26 (merge time)　Source: main repository commit log, PR #11885

The mainline had been idle for about two days after the evening of 10-25. Between 19:20 and 19:26 on 10-27, 4 commits were merged consecutively within six minutes, all by the same maintainer, touching dfs, net/sal, libc/posix, and drivers/serial, for a total of 46 insertions and 6 deletions. Three of the four have an author date of 10-18, indicating a batch landing of backlogged PRs rather than code written that day. This was also the only mainline change in this issue.

1.2 dfs UTIME_OMIT check is inverted, overwriting timestamps explicitly marked to be left unchanged (10-27)

Date: 2026-10-27 19:20 (merge time)　Source: PR #11885

In the dfs_v2 utimensat implementation, the branch condition for UTIME_OMIT was inverted. Per POSIX semantics, a timestamp field set to UTIME_OMIT must be preserved as-is; the original code continued to write mtime under the negated condition, effectively overwriting a timestamp the caller explicitly asked not to touch while also dropping a valid timestamp. The fix changes a single comparison operator, with a net change of one line in dfs_posix.c. This kind of semantic deviation is invisible to test cases that only verify “setting the time succeeds,” making it a latent defect.

1.3 RAW socket protocol validation switched to a whitelist, protocols like ICMP no longer blocked (10-27)

Date: 2026-10-27 19:21 (merge time)　Source: PR #11888

When the SAL layer created a RAW socket, it previously gated on a single equality check of whether protocol equals IPPROTO_RAW, so every protocol other than IPPROTO_RAW was rejected, blocking scenarios such as ICMP that require raw sockets. This change replaces it with VALID_PROTOCOL whitelist validation, while the existing domain and type checks remain unchanged. The same PR also includes a purely cosmetic macro-alignment change with no functional impact. The change lands in sal_socket.c, with a net change of one line.

1.4 poll now resolves fops via dfs_file, compatible with both DFSv1 and DFSv2 (10-27)

Date: 2026-10-27 19:21 (merge time)　Source: PR #11887

poll previously fetched fops directly from the vnode, which under DFSv2 returned a structure that is not the same set as the actual file operation table, effectively attaching poll to the wrong dispatch entry point. This change resolves fops through dfs_file, retains the DFSv1 compatibility path, and adds a null-pointer check before dereferencing. The change is concentrated in poll.c, with 14 insertions and 2 deletions. This is one of the adaptation issues left over from the DFSv2 switch, and the fix leans conservative rather than a rewrite.

1.5 8250 earlycon starts reading reg-io-width, wide-register platforms no longer accessed byte-wise (10-27)

Date: 2026-10-27 19:26 (merge time)　Source: PR #11889

The early serial console previously accessed registers at byte width regardless of what the device tree specified. This change reads reg-shift and reg-io-width from the device tree, selects the MMIO access width based on reg-io-width, and rejects unsupported widths outright during early initialization, while other compatible strings still take the generic 8250 path. The change lands in early.c, with 30 insertions and 2 deletions. For platforms with non-default register stride or bit width, early log misalignment and garbled output now have a unified fix, eliminating the need for per-board code changes.

1.6 Release side still at v5.3.0; mainline bumped to v5.4.0 on 10-17 but no tag yet (10-27)

Date: 2026-10-27 (verification time)　Source: release records, tag list

Neither releases nor tags show any new entries within 24 hours; the latest release remains v5.3.0 from 9-10. The mainline has a commit on 10-17 bumping the version number to v5.4.0, indicating the development branch version has already moved to 5.4.0, but the repository has no corresponding tag or release record yet. Based on the pattern of previous years, official releases are typically announced only after tagging, so this issue does not yet constitute a release signal.

II. Merge Queue and Security Follow-up

2.1 signalfd close path gains slot bounds checking and listener unbinding (10-27)

Date: 2026-10-27 08:34 (created)　Source: PR #11935

In response to the signalfd global counter overflow recorded in yesterday’s briefing, the community submitted a fix proposal the same day. The approach replaces the global index counter with a context-localized, bounded slot lookup; on signalfd_close, it walks the registered lwp instances, unlinks the corresponding notification node from the list and frees the listener structure, then resets the file context pointer and zeroes the table entry to prevent stale pointers from being dereferenced again. The proposal adds a net 41 lines in a single file and is currently mergeable. Less than a day passed between the original issue and the proposal — a rare response speed.

2.2 Kernel-wide clang-format 22.1.8 formatting proposal under review, touching 75 files (10-27)

Date: 2026-10-27 17:42 (created)　Source: PR #11936

A proposal pinning clang-format to version 22.1.8 and reformatting the kernel source has entered the queue, touching 75 files with 3922 insertions and 3040 deletions. This follows the style-rule refinements merged on 10-23, taking the next step of applying those rules to the full kernel codebase. Such proposals contain no functional changes themselves, but they create large-scale textual conflicts with other kernel proposals under review, so the merge order needs maintainer coordination — typically it lands only after a wave of feature proposals has cleared.

2.3 ESP32-S3-DevKitM-1 proposal refreshed at the end of the window, now at 295371 cumulative insertions (10-28)

Date: 2026-10-28 07:38 (last updated)　Source: PR #11844

The ESP32-S3-DevKitM-1 BSP proposal was refreshed again at the end of the window, now at 295371 cumulative insertions and 22825 deletions — the largest item in the review queue. Created on 9-28, the proposal has now been open for a full month, and the repeated refreshes indicate the author is still working through review comments. At this scale, review attention typically falls on whether to import the vendor SDK wholesale, along with binary libraries and license terms, rather than merely whether the functionality works — which is why it has lingered in the queue so long.

2.4 Four RT-Thread Smart architecture proposals still in queue, all submitted 10-18 (10-27)

Date: 2026-10-27 (last updated)　Source: PR #11894, PR #11896

Four proposals related to RT-Thread Smart remain in the queue, all created on 10-18 as part of the same batch of work: splitting lwp syscall implementations by subsystem (12223 insertions, 11638 deletions), enabling Smart on qemu-virt64 (1263 insertions, 1075 deletions), fixing MMU and user-mode page-fault handling in libcpu/riscv (105 insertions, 8 deletions), and adding K230D support to k230 (587 insertions, 511 deletions). All four were updated between 10-23 and 10-27, indicating the author is iterating on review feedback, but none has been merged yet. The same 10-18 batch began landing today; the Smart group is likely the next batch.

III. Toolchain, Official, and Community

3.1 Community long-form post breaks down Env v2.0.3: commands consolidated into rt-env, new plugin system and WebUI added (10-27)

Date: 2026-10-27 11:01 (publish time)　Source: Community article: Full record of using Env Tool v2.0.3

The community published a complete usage record of the Env toolchain upgrade to v2.0.3. The article notes that Env has undergone an architectural refactor since 2.0: the unified command entry point is now rt-env, with menuconfig, pkgs, and scons no longer standalone subcommands, and new additions of sdk, plugin, webui, and system; all dependencies run inside a Python virtual environment, isolated from the system Python; a new plugin system and WebUI allow toolchain, SDK, and package management to be completed in the browser; the trade-off is that every terminal must be activated before commands can be used. The version combination the author tested is rt-env 2.0.3 with SCons 4.11.1 and kconfiglib 14.1.0. The article specifically flags a compatibility red line: Env 2.0 fully supports versions above 5.1.0 and the master branch, while projects on 5.1.0 and earlier must remain on the 1.5.x release package.

3.2 Official news API shows no new items in 24 hours; latest issue still stuck at June 17 (10-27)

Date: 2026-10-27 (verification time)　Source: Official news API

After a page-by-page check of the official news API, the latest entries remain the two from 2026-06-17, covering the Abu Dhabi Investment Forum and the Fourth Software Innovation Development Conference respectively, with no new items in either the 24-hour or 7-day window. By contrast, the community article API had one new post in this window (see 3.1), indicating that official external releases and community submissions travel through two independent channels, with the official side having been at low frequency for four consecutive months. The daily briefing’s routine content is therefore effectively carried by GitHub and the community side.

3.3 EE Times China’s three-way RTOS comparison lists RT-Thread as preferred for low-power and consumer scenarios (10-27)

Date: 2026-10-27 22:27 (publish time)　Source: EE Times China: In-depth three-way RTOS comparison

A column under EE Times China published a three-way RTOS comparison, contrasting FreeRTOS, RT-Thread, and Zephyr across six dimensions: real-time performance, memory footprint, ecosystem, community, AI support, and migration cost. The article positions RT-Thread as an Apache 2.0 lightweight all-in-one package, with a kernel of 3KB Flash plus 1.2KB RAM, and built-in POSIX, GUI, file system, and AI packages; its low-power components save 15% to 20% more power than the other two in Stop mode. The selection decision table assigns ultra-low-cost projects under 64KB RAM, as well as consumer electronics and wearables, to RT-Thread. The article also acknowledges that Zephyr is stronger in multi-protocol, multi-board, and long-term maintenance scenarios, concluding with a three-way split rather than a one-sided verdict; the data is cited from each project’s official documentation and third-party benchmarks, with no visible signs of vendor sponsorship.

IV. Ecosystem and Vendor Adaptation

4.1 Arm China WTCC 2026 Workshop: RT-Thread Focuses on Embodied Intelligence Projects, Breaks Out Three Agent Approaches (10-27)

Date: 2026-10-23 (event) / 2026-10-27 14:13 (report)　Source: Mydrivers report, 21ic same article

Arm China held a small embodied intelligence solutions workshop during the Shanghai West Bund International Tech Consumer Carnival (WTCC 2026), with Rockchip, Radxa, and RT-Thread participating as partners of the Open Source AIOS Alliance. RT-Thread senior AI engineer Zheng Wei presented three Agent approaches—robot dog, HyperCar, and Codex orchestration—explaining how agents move from conversation to execution, and made clear that RT-Thread will focus on embodied intelligence projects going forward, working with Arm China to map Agent capabilities onto real applications. Rockchip noted that LLM inference is shifting from compute-intensive to memory-intensive, with bandwidth becoming the edge bottleneck, and proposed decoupling compute from control using a host SoC plus an AI co-processor. Arm China also released two deliverables at the event: JishuFriends and JishuBuddy.

4.2 Kitchen Flame Detection Prototype on K230 Opens Source Code and Reproduction Records, Built on RT-Thread Smart (10-27)

Date: 2026-10-27 19:19 (push time)　Source: SRTP-K230-Fire-Guardian

A kitchen flame detection prototype built on K230 and RT-Thread Smart has open-sourced its code, covering multi-sensor fusion, adaptive linkage strategies, reproduction records, and model evaluation; the repository also includes reference materials so third parties can rerun it following the records. The project describes itself as a research prototype rather than a product. This is a rare public project on the ecosystem side this issue that applies RT-Thread Smart in a real perception loop, offering a useful reference for assessing Smart’s usability in vision scenarios and showing that community-side developers beyond board vendors are starting to work with Smart.

4.3 Handwritten Symbol Recognition with RT-Thread, LVGL, and TFLM on the CH32H417 Dual-Core RISC-V (10-27)

Date: 2026-10-27 08:41 (push time)　Source: ch32h417-tinyml-symbol-recognition

A handwritten symbol recognition project on WCH’s CH32H417 dual-core RISC-V platform, using a stack of RT-Thread, LVGL, and TFLM. The pipeline covers OV5640 capture, fixed-point image preprocessing, INT8 quantized model inference, and UI display, with a separate toolchain on the training side. The project is newly created and has no traction yet; its value lies in laying out a complete lightweight vision pipeline on RISC-V clearly, from data collection to quantized deployment, making it a reusable example for teams preparing to do edge inference on domestic RISC-V MCUs.

V. Industry Applications and Embodied Intelligence

5.1 AIoT Competition Entry Turns to Industry: AI Monitoring of BLDC Motor Defects on Titan Board (10-27)

Date: 2026-10-27 11:50 (media repost)　Source: EET-China: BLDC Motor Defect AI Monitoring System

EET-China reposted an entry from the 2026 RT-Thread AIoT Competition: AI monitoring of BLDC motor operating-state defects on the Renesas Titan Board. The hardware is the RA8P1—a 1GHz Cortex-M85 plus a 250MHz Cortex-M33 dual-core, with an Ethos-U55 NPU, delivering 256 GOPS on a single chip. The system collects three-phase current, bus voltage, and inverter temperature, preprocesses on the M85 core, and feeds a lightweight CNN deployed on the NPU for real-time defect recognition; on the software side, RT-Thread runs three core threads for UI display, data communication, and AI inference, with an LVGL-based UI and the model converted from TFLite to C code via RUHMI for integration. The approach emphasizes adding no extra sensors and analyzing state solely from drive current and voltage—a low-cost predictive maintenance path.

5.2 This Issue’s Thread: Chip Vendors, Board Vendors, and the OS Side Share a Stage Under the Alliance Framework, All Pointing to Embodied Intelligence (10-27)

Date: 2026-10-27 (this issue)　Source: Mydrivers report, SRTP-K230-Fire-Guardian

Viewed side by side, the three threads in this issue point in a fairly consistent direction. On the upstream IP and chip side, Arm China brings Rockchip, Radxa, and RT-Thread to the same table through the Open Source AIOS Alliance, discussing compute decoupling and edge bandwidth bottlenecks; on the OS side, RT-Thread makes embodied intelligence projects a priority, talking about agents moving from conversation to execution; on the community and board side, the K230 and CH32H417 projects apply Smart and lightweight inference respectively to concrete tasks like flame detection and symbol recognition. All three are converging on “running a complete perception-and-decision loop on small devices” rather than each expanding its own feature list.

VI. Summary and Trend Observations

6.1 This Issue’s Trend: Merges Resume, Fixes Concentrate on Boundary Validation Rather Than New Features

Date: 2026-10-28 (this issue)　Source: Main repository commit log

The most notable item this issue is the recovery of the merge cadence and its composition. Of the four merges, three have author dates of 10-18, meaning that batch of proposals sat in review for nine days and is now landing one by one; four more Smart-related proposals remain queued in the same batch, so the mainline is expected to stay dominated by them over the coming days. The four merges share a consistent technical orientation: filesystem timestamp semantics, socket protocol whitelists, poll dispatch table sources, and UART register widths—all boundary validation and semantic alignment, with no new features. These four areas rarely report errors in normal operation but are hard to localize once something goes wrong; fixing them together in one batch is a positive for long-term stability.

6.2 Limitations and Deduplication Notes

Date: 2026-10-28 (this issue)　Source: Official news API

  • The official news API has had no new entries since June; all three media-side items this issue were first found via aggregated search and then verified against the original sites. Of these, the two EET-China pieces are a column roundup and a competition entry repost, respectively—neither an official RT-Thread release.
  • The signalfd out-of-bounds issue (Issue #11933) recorded in yesterday’s daily report appears this issue as proposal #11935; it is included as a new development and not counted as a duplicate entry.
  • Not included: the 5 commits in zlgopen/awtk that day are not directly related to RT-Thread; Nuclei’s nuclei-sdk has a pushed_at within the past week, but its actual latest commit is dated June 10—a branch or tag change, not a code update.
  • All 7 HN Algolia hits this issue were false string matches on “threads” and “Threads” and were removed as a batch; Google News English side had no relevant articles in the window.

Appendix: Source Verification Table and Source List

Source Verification Table

Source Verification Result
Official news API (rt_org) No new items in 24 hours; latest entry remains at 2026-06-17
Community article API (article) 1 new item in the window: Env v2.0.3 usage notes (10-27 11:01)
Main repo commits 4 merges in the window (10-27 19:20 to 19:26), following roughly two days of silence
Main repo PRs 2 new (#11935, #11936); 10 open proposals updated in this window
Main repo releases / tags No new items; latest remains v5.3.0 (2026-09-10)
Main repo Issues No new issues; two old 2023 issues saw activity
RT-Thread/packages No pushes; latest commit 2026-10-23
vscode-rt-smart No pushes; latest commit 2026-10-26
RT-Thread/sdk No pushes; latest commit 2026-10-25
pkgs-test No pushes; latest commit 2026-10-23
topic:rt-thread ecosystem search 4 repos pushed in the window: SRTP-K230 and CH32H417 included; awtk and nuclei-sdk excluded
Google News Chinese (seven queries) 4 hits in the window; 3 included, 1 deduplicated as same-story
Google News English No relevant hits in the window
HN Algolia All 7 were false matches on threads and Threads; entire batch excluded

Full Source List