Research window: 2026-11-18 08:00 ~ 2026-11-19 08:00 (Beijing time, 24 hours)

Sources: RT-Thread official news API (category rt_org), GitHub main repository commits and PRs, releases and tags. Also includes pushes across all RT-Thread org repos and the package index repo packages. Supplementary channels are the Windows tools repo env-windows, active repos under topic:rt-thread, Google News (via proxy), and Hacker News.


Index for This Issue

  • I. Device Layer
    • A 2021 issue, closed by changing the encoding (11-18)
    • Where the codes collide: fcntl’s 1 through 4 share values with device commands (11-18)
    • Fourteen files changed along with it, spanning BSPs and rt-link (11-18)
  • II. Testing and CI
    • memheap test cases silently dropped on BSPs without MEMHEAP (11-18)
    • Mixed path separators in the utest documentation (11-18)
    • Three commits in the early hours, one landed the same day (11-18)
  • III. Packages and Ecosystem
    • AI Agent package adds A2A channel and layered memory (11-18)
  • IV. Official and Community
    • Official news API sees no new items for the ninth consecutive issue (11-19)
    • No new releases, no new tags (11-19)
  • V. Review Queue
    • RH850 port grows to 52 files within a week (11-18)
    • The other three touched only on timestamps (11-18)
    • No new proposals or issues in the window (11-19)
  • VI. Summary and Trends
    • Today’s weight rests on a single proposal
    • Silent dropping is the hardest gap to spot on the testing side
    • After command values shift, out-of-repo code must verify itself
    • The second change to the AI Agent package

I. Device Layer: Redrawing the Command Space

1.1 A 2021 Issue, Closed by Changing the Encoding (11-18)

Date: 2026-11-18 09:11 (merged). Source: #11881.

Issue #5073 was filed on September 17, 2021. Its title says that serial ports going through the POSIX ioctl interface cannot set low-level parameters and never reach the underlying config, with the cause traced to “the macro-defined value being 3.” The issue remains open to this day.

PR #11881 was opened on October 17, 2026. At 9:11 a.m. on November 18, Rbb666 merged it. That is 5 years and 2 months from report to merge, and 32 days from proposal to merge.

Commit 629a7970. The scope is 14 files, 80 lines added, 57 lines removed. The approach is not to patch a single driver but to replace the numbering scheme for device control commands wholesale.

1.2 Where the Collision Sits: fcntl’s 1 to 4 Share Values with Device Commands (11-18)

Date: 2026-11-18 (PR body). Source: #5073.

The PR body lays the conflict out in a table. On the POSIX side, F_GETFD, F_SETFD, F_GETFL, and F_SETFL are 1, 2, 3, and 4 respectively. In the old definitions, RT_DEVICE_CTRL_RESUME, SUSPEND, CONFIG, and CLOSE are 0x01, 0x02, 0x03, and 0x04.

Both command sets share a single int argument. F_GETFL collides with CONFIG at 3, and F_SETFL collides with CLOSE at 4.

The path is spelled out as well. fcntl goes through dfs_file_ioctl and dfs_file_fcntl to fops->ioctl. When a driver matches by numeric value, it treats F_GETFL as CONFIG. This is precisely the mechanical cause of “cannot set parameters” in issue #5073.

The new encoding moves device classes into the high byte. The base macro is RT_DEVICE_CTRL_BASE(Type). It is the class number plus one, times 0x100. The high byte holds the class number, the low byte the index within the class.

The nine generic commands switch to BASE(Unknown) + 1..9, landing at 0x2001 through 0x2009. This is fully disjoint from POSIX’s 0 to 4.

The old RT_DEVICE_CTRL_MASK = 0x1f looked only at the low 5 bits and is incompatible with the new encoding. The author grepped the whole repo, found no callers, and deleted it. It is replaced by RT_DEVICE_CTRL_CLASS_MASK = 0xff00 and RT_DEVICE_CTRL_INDEX_MASK = 0x00ff.

One detail is worth noting. The “verified BSPs and configs” field in the PR is empty, and the .config and action links are not filled in either.

Date: 2026-11-18 (commit diff surface). Source: 629a7970.

Of the files that follow the change, eight are efm32 board-level drivers. acmp, adc, ethernet, iic, leuart, sdcard, timer, and usart each change three to four lines of command values. hk32f030c8-mini’s rtconfig.py changes by one line.

The header-side changes are more concentrated. In dev_pin.h, PIN_CONFIG_BIAS_BUS_HOLD was originally the first enum entry with value 0, and is now RT_DEVICE_CTRL_BASE(Pin) + 1.

dev_serial_v2.h turns three bare macros into an enum. They are RT_HW_SERIAL_CTRL_GETC, PUTC, and GET_DMA_PING_BUF. Values start at RT_DEVICE_CTRL_BASE(Char) + 0x40, with a note that 0x40 through 0x4f are reserved for hardware serial control.

Two rt-link files follow as well. include/rtdef.h is the largest single change, with 34 lines added and 16 removed. A line is added to the version comment: 2026-11-07, CXSforHPU, class-encode RT_DEVICE_CTRL_* commands.

Assessment: The nine generic commands move from 0x01–0x09 to 0x2001–0x2009. Out-of-tree BSPs or packages that hard-code the old values will take the wrong branch after upgrading. This is based on the commit syncing only 14 files inside the repo, leaving no compatibility macros, and the verified-BSP field being empty.

II. Tests and CI Wrap-Up

2.1 memheap Cases Silently Dropped on BSPs Without MEMHEAP (11-18)

Date: 2026-11-18 17:29 (merged). Source: #12066.

The commit message spells out the causal chain. RT_UTEST_MEMHEAP depends on RT_USING_MEMHEAP. mem.cfg does not select it. So the memheap test group was not compiled in on BSPs without that config. The old mem.conf had it selected.

The fix adds two lines to .github/utest/configs/kernel/mem.cfg. PR #12066 was opened at 3:13 a.m. that day and merged at 17:29, a 14-hour gap.

This kind of gap does not fail the build, nor does it turn tests red. The report shows a pass; the dropped cases simply do not appear in the statistics.

2.2 Mixed Path Separators in the utest Docs (11-18)

Date: 2026-11-18 10:01 (merged). Source: ebd32d7a.

components/utilities/utest/README.md changes two lines, unifying mixed path separators. The author is sexenteen; written on November 16, merged on the morning of November 18. A documentation-level change with no behavioral impact.

2.3 Three Overnight PRs, One Landed the Same Day (11-18)

Date: 2026-11-18 (PR status). Source: #12064.

The three overnight PRs listed yesterday have taken different paths today. #12066 is merged. #12064 and #12065 remain open, last updated at 19:08 and 20:25 respectively.

All three come from parthitce. #12065’s body states the test environment: kernel_basic on RH850 qemu-rh850-u2c. Under -O2, stack corruption crashes rt_memmove; after the fix, all 22 test cases pass.

The defect in the test case itself is that it reads and writes 11 bytes against two arrays of 6-byte source and 10-byte destination.

III. Packages and Ecosystem

3.1 AI Agent Package Adds A2A Channel and Tiered Memory (11-18)

Date: 2026-11-18 09:45 (commit). Source: 6595550d.

The package index repo received a commit touching only ai/Agent/Kconfig, with 378 lines added and 212 removed. The commit message lists three things.

First, add a2a channel support. Second, add switches, capacity, and persistence options for tiered memory. Third, when persistence is enabled, automatically select DFS and the POSIX file interface at 4.1 or above, so users need not go back and add components.

The package itself, ai/Agent, describes itself as “rtthread lightweight Agent package.” Keywords are ai, Agent, llm, toolcall. The repo points to the author’s own CXSforHPU/Agent, depending on cJSON, webclient, mbedTLS, and ulog.

This is the second change to this menu in two days. On November 16 at 11:22, #2050 “Complete Agent config items, reorder menu by purpose” was merged, also touching this Kconfig. The author, CXSforHPU, also merged the command-encoding commit in 1.1 to the main repo today.

IV. Official and Community Activity

4.1 Ninth Consecutive Period with No New Official News (11-19)

Date: 2026-11-19 (API check). Source: API.

The API returns five items. The latest is still “From OS Foundation to AI Deployment: RT-Thread’s Full-Stack Capabilities Anchor Industrial Intelligent Upgrade” from November 6 at 15:54. Counting from November 7, this is the ninth consecutive period with zero additions.

4.2 No New Release, No New Tags (11-19)

Date: 2026-11-19 (release and tag check). Source: releases.

The latest release is v5.3.0, published on September 10. The tag list matches, with no additions in the window.

The version comment on the main branch was already recorded as v5.4.0 on October 17. The reporter of issue #12063 also writes master as v5.4.0-dev. The official release remains 5.3.0, one development cycle apart.

V. Review Queue

5.1 RH850 Port Grows to 52 Files in a Week (11-18)

Date: 2026-11-18 20:15 (last update). Source: #11998.

The Renesas RH850 G4KH port and qemu-rh850-u2c board package were opened on November 6. Yesterday it was 48 files and 3921 lines added. Today it is 52 files and 4122 lines added—4 files and 201 lines in one day.

The content falls into three parts. libcpu/rh850 uses TRAP0 for thread switching, including nested interrupts, FPU context, and fault dumps. bsp/renesas/rh850 has startup code, INTC, OSTM, and RLIN3 serial, with serial supporting only serial V1. CI builds both hard-float and soft-float variants, plus kernel utest on QEMU, with SHA-256 verification of the toolchain and QEMU packages before extraction.

The author’s stated test environment is toolchain v1.2.0 (GCC 14.3, picolibc) and qemu-rh850 v1.2.0.

Assessment: This PR is still adding content, not waiting on review. This is based on file count and line count both rising within a day while the review-side comment count shows no corresponding growth.

5.2 The Other Three Only Moved on Timestamps (11-18)

Date: 2026-11-18 19:26 (last update). Source: #11842.

HUST CSE’s fix for the devtmpfs overlong-path component stack overflow is still 5 commits, 4 files, 314 lines added, 8 removed. Consistent with yesterday’s check, last updated at 19:26. For #12064 and #12065, see 2.3.

All three only refreshed their update times; no new commits came in.

5.3 No New PRs or Issues in the Window (11-19)

Date: 2026-11-19 (PR and issue counts). Source: PR list.

In the window, 0 new PRs and 0 new issues. Open PRs: 138, down from 141 last period. Open issues: 341, flat versus last period.

VI. Summary and Trend Observations

6.1 Today’s Weight Rests on a Single PR

The window contains three commits, 16 files total, 84 lines added, 59 removed. Of these, 629a7970 alone accounts for 14 files, 80 added, 57 removed. The other two each touch one file and two lines each.

Assessment: Today is a redrawing of the device command space, not patch stacking. This is based on the changes concentrating in include/rtdef.h and four device headers, with the BSP side only following the value changes and no logic changes.

6.2 Silent Dropping Is the Hardest Test-Side Gap to Catch

The memheap test group did not fail to compile; it was never compiled in. The build passes, the tests pass. Nothing looks wrong in the report.

Assessment: This class of problem can only be found by comparing configs, not by running tests. This is based on the mechanism stated in the commit message: RT_UTEST_MEMHEAP depends on RT_USING_MEMHEAP, and when mem.cfg does not select it the whole group is skipped, whereas the old mem.conf did select it.

6.3 After the Command Values Shift, Out-of-Tree Code Must Check Itself

Assessment: The nine generic commands move from 0x01–0x09 to 0x2001–0x2009; out-of-tree BSPs and packages that hard-code the old values will take the wrong branch after upgrading. This is based on the commit syncing only 14 files inside the repo, leaving no compatibility macros, and the verified-BSP field being empty.

6.4 The Second Change to the AI Agent Package

Menu reordered on November 16, A2A channel and tiered memory added on November 18. Both touched only Kconfig.

Assessment: This package’s configuration surface is expanding toward “protocol access + memory management,” no longer just feature switches. This is based on the second commit adding a2a channel plus memory capacity and persistence options, whereas the first only reordered the menu by purpose.

Appendix: Source Verification Table

Source Verification Result
RT-Thread official news API Five items, latest 11-06 15:54, ninth consecutive issue with zero new items
Official news page Client-side rendered, items cannot be scraped; API is authoritative
Main repo commits 3 in window, 16 files, +84 / -59 lines
PRs merged in window 2: #11881 (command encoding), #12066 (mem.cfg)
PRs opened in window 0
Issues opened in window 0
Open PRs / issues 138 / 341 (previous issue 141 / 341)
releases and tags Latest remains v5.3.0 (2026-09-10), no new items in window
Package index repo packages 1 in window (ai/Agent config, +378/-212)
Windows tools repo env-windows Last push 11-16 17:14, zero commits in window
Other RT-Thread org repos No pushes in window; top two by push recency are rt-thread, packages
topic:rt-thread active repos 5 repos; only zlgopen/awtk pushed two commits on 11-18 16:42 (bitmap font loading, framebuffer stride), general GUI improvements, no specific RT-Thread tie-in
Google News Chinese Zero items in window; most recent is a contest grand-prize project share from 38.6 hours ago (covered yesterday); zero hits for 睿赛德 and 睿擎派 over three days
Google News English Zero hits in three-day and seven-day windows; bare queries work normally, ruling out channel failure
HN Algolia Both candidates were literal “thread” false matches, excluded
Security advisories No RT-Thread entries, no new IDs this issue

Full Source List