⚠️ Editorial note: The open source ecosystem in China operates under a distinct institutional framework — characterized by state-led initiatives, intranet-like boundaries, and top-down governance. Readers should be aware that this context differs from the community-driven open source model common in other regions. The term “open source” as used in Chinese media may refer to practices that diverge from the conventional definition.
China Open Source Daily — 2026-11-05
🏗️ Institutional Change — Where the Work Happens: An AtomGit Landing Page, a GitHub Mirror, and 400 Commits of openEuler’s Kernel
1. The two artefacts, and why they are worth reading together. On 2026-10-20 The Register published an assessment of openEuler at Open Source Summit Europe in Prague: the only dedicated Linux-distro booths at the summit belonged to Chinese projects, openEuler’s was the larger, and the piece’s conclusion was that “half a decade after it was cut off from its upstream sources, the openEuler platform is thriving and growing, even if few in the western world pay any attention at all.”
Four days later, a different and arguably more informative artefact was written. The canonical home of the openEuler kernel repository moved from Gitee to AtomGit. Gitee now carries the notice in both languages — “This project has been migrated to AtomGit” / 本项目已迁移至AtomGit — with the linked destination atomgit.com/openeuler/kernel. Gitee still shows the old counter: 1,245,880 commits on OLK-6.6, 1,657 stars, 1,868 forks, 1888 tags.
On GitHub there is a repository called openeuler-mirror/kernel. It is still alive: last push 2026-11-04T04:40:03Z, default branch OLK-6.6, 59 stars, 41 forks, 0 open issues.
2. What each surface says about participation. The numbers are stated plainly because the contrast is the finding:
| Surface | Stars | Forks | Open issues | Pull requests | Members |
|---|---|---|---|---|---|
AtomGit openeuler/kernel (canonical) | 590 | 676 | 1,849 | 1,486 | 203 |
GitHub openeuler-mirror/kernel (mirror) | 59 | 41 | 0 | 0 | — |
GitHub’s issues and pull-request endpoints on that repository return empty lists, and GitHub search for repo:openeuler-mirror/kernel is:pr returns 0. This is not moderation and it is not an outage: it is the normal condition of a mirror. The conversations are somewhere else.
3. What the 400 most recent mirror commits contain — the one part of this that can be measured mechanically. GitHub’s API is the only one of the two forges that answers without a session, so the sample below is drawn from it and labelled as such: the 400 most recent commits on OLK-6.6, spanning 2026-07-22 → 2026-11-04. Two figures are worth carrying:
Of the most recent 100 commits, 59 are individual patches rather than bot merges — real human work landing on a branch. 45 of those 100 are authored in the foreign mainline (
kuniyu@google.com,edumazet@google.com,idosch@nvidia.com,jirislaby@kernel.org,kbusch@kernel.org,mhiramat@kernel.org,qiuxu.zhuo@intel.com,vishal.badole@amd.com, …). The remaining 55 in that window are authored byopeneuler-ci-bot <george@openeuler.sh>.Those bot commits are not noise; they are a translation. A representative message, taken verbatim from the mirror:
!29993 [CVEAgent@OLK-6.6] Fix CVE-2026-98807From: @zhangxiaoxvReviewed-by: Lu Jialin <lujialin4@huawei.com>Reviewed-by: Tengda Wu <wutengda2@huawei.com>Signed-off-by: Tengda Wu <wutengda2@huawei.com>The
!29993prefix is an AtomGit merge-request number. TheFrom: @zhangxiaoxvis an AtomGit account handle. The commit landed on the public Git tree with a Huawei reviewer’s sign-off attached to it, because the review happened on the other forge. In the same 400-commit sample, commits carrying reviewer sign-offs come overwhelmingly from Huawei addresses (luijialin4@,wutengda2@,liuyongqiang13@), withkylinos.cnappearing as the only other corporate reviewer domain visible in the sample.The content of the patches is mainline inclusion, not original development. A commit visible on the AtomGit landing page is explicit about its own provenance:
neighbour: Enforce min/max to NDTPA_INTERVAL_PROBE_TIME_MS.—mainline inclusion from mainline-v7.3-rc4 commit 6d79b223ec44ada58ad37db42f539b60985a7722,CVE: CVE-2026-98317,Signed-off-by: Zhang Changzhong <zhangchangzhong@huawei.com>, ending inConflicts:against two named upstream commits.That is a backport — upstream work carried back onto the distribution’s kernel, with the conflicts and the CVE written down. The tags on the mirror confirm the same relationship from the other direction:
v7.3(tagger date 2026-10-18),v7.0(2026-04-12),v6.18.56(2026-10-11),v6.18.48(2026-08-28), all carrying Linus Torvalds’s signed release messages.
4. The other side of the ledger, stated at the strength the sources support. Two sets of figures circulate for this project and they are not the same kind of number.
- Participation, from the community’s own monthly report (2026-09-30): 29,000-plus developers (超过2.9万名), 2,199 member organisations, 112 SIGs, 340,800 cumulative PRs, 167,500 issues, 6.2 million comments, and more than 10 million cumulative community users. The openEuler 26.09 innovation release, published 2026-09-30, credits 834 developers for the release — a specific, attributable number rather than a cumulative one.
- Adoption, from executive statements: openEuler installs passed 20 million (Huawei deputy chairman Wang Tao, Huawei Connect, 2026-09-17/19); 2025 added 4.48 million new installs with the openEuler line at 57.3 percent of the 2025 incremental server-OS market (Frost & Sullivan / LeadLeo white paper, as reported); the community’s April 2026 Developer Day put cumulative installs at 16 million and share at 57.3 percent. Huawei’s own open-source page still carries an older 16 million install / 5.8 million downloads / 2,100-plus members figure, alongside “Linux Kernel 6.10 代码贡献排名 NO.1”. These are self-reported and not independently verifiable; they are recorded as reported, and the divergence between them is itself a fact about how this ecosystem measures itself.
Institutional Analysis & Commentary:
The interesting distinction is not “real or fake.” It is location.
Every serious judgement of Chinese open source collapses when it is asked to be a verdict on authenticity, and this repository is a clean example of why. The kernel is GPL-2.0 WITH Linux-syscall-note (COPYING, SPDX-License-Identifier), the full source is published, the tags track real mainline releases, the commit messages occasionally admit their own conflicts, and the community publishes a monthly operational report with SIG counts and CVE tallies. None of that is theatre. Meanwhile the same project’s own landing page literally ranks a Huawei code-contribution claim beside a project that was moved off Huawei’s direct control in 2021 — the donation to OpenAtom did not make the code stop flowing from one address.
The honest description is narrower and more useful: this is an open downstream. openEuler’s kernel is a maintained fork of Linux, and the work in its sample is backporting, CVE response, and hardware-enablement. That is the same category of activity Rocky Linux and AlmaLinux do, and it is legitimate engineering. The difference that matters is not the licensing and not the openness of the source. It is where the decision surface sits: 1,486 pull requests and 1,849 issues on AtomGit; zero and zero on GitHub; a Gitee repository frozen with a migration notice and a 1.25-million-commit counter that no longer increments.
Williamson’s layers separate cleanly here, and the separation is the argument.
At L2, the institutional environment — licence, publication obligation, the foundation’s project lifecycle — the commitment is real. openEuler reached the OpenAtom Graduation milestone in November 2025, a judgement the foundation defines as comprehensive maturity in technical iteration, community governance and ecosystem development. Graduation is a procedural fact, checkable and not rhetorical.
At L3, governance — who decides, who reviews, who merges — the mirror’s own metadata says it plainly. The review happened upstream-in-China at !29993; the GitHub copy carries the sign-off as an artefact. A mirror can reproduce a merge. It cannot reproduce a merge request. And the only forge where a stranger outside China could read the review conversation is the one whose issue list is empty by design.
At L4, resource allocation, the arrangement works: the procurement market, the OSVs, the 20 million installs. This is the layer where the project is an unqualified success, and pretending otherwise would be the mirror image of the boosterism this briefing exists to counter.
What the mirror actually proves, and it is more than nothing.
A mirror is a costly commitment to verifiability, and it should be read as one. Publishing a Git tree nightly means anyone can git clone, diff a CVE fix, check whether a security patch was backported correctly, and — importantly — audit the project without asking permission. That property is what open source licences are for, and openEuler has it in a way the model-weight ecosystem discussed in yesterday’s briefing does not: you can read every line of the kernel, and you can read the CVE numbering (CVE-2026-98807, CVE-2026-98347, CVE-2026-98855) attached to each fix.
The gap is between a mirror and a place where work happens. One-way publication is a credible commitment to stay open, which is genuinely valuable and underpriced in Western commentary. Two-way traffic is a commitment to share control, which is a different and far more expensive thing. openEuler has spent the first and not the second, and it has done so in the middle of a long march upstream, not in place of it. The community’s own materials list named contributions to the Linux kernel — architecture support, ACPI, memory management, filesystems, media, documentation — from members including Huawei, Loongson, Kylinsoft, UnionTech, Phytium, China Telecom and China Mobile. This is not a project that refuses upstream. It is a project that runs its own forge, upstreams what it can, and backports what it cannot.
Where the arrangement becomes a risk is the mirror’s one condition: that nothing changes.
A one-way mirror is safe while publishing stays free and automatic, and it is a synchronisation dependency without a synchronisation obligation. Nobody has promised anyone a nightly push. Consider the sequence already visible in the record: the canonical host moved from Gitee to AtomGit at some point before 2026-11-04, and the 1.25-million-commit Gitee counter went quiet. Both were internal decisions made by the project in its own interest, and both were defensible. Neither required notice, because the mirror has no contract attached to it. The mirror is not the institution; it is a courtesy, and a courtesy is exactly as strong as the current relationship.
And that is the fork below the surface, measured honestly and without alarm. The two kernels — 6.6 in the LTS line and 6.18 in the 26.09 innovation release — are both selectively backported, vendor-reviewed ports carrying named conflicts against the mainline. The mirror’s non-upstream kernel work referenced by LWN for the RISC-V port (a six-year-old architecture line, still outside mainline) is the extreme case, not the general one. Today the divergence is one a determined engineer can close in an afternoon with git log. In five years of security backports applied at a thousand commits a month under CVE pressure, from a forge that publishes to GitHub as a courtesy, that afternoon becomes a week. That is how forks arrive — never as an announcement, always as an accumulation nobody had to approve.
The judgement, stated as a judgement rather than a worry.
Is openEuler open source? On the licence, the source, the graduation milestone, the monthly operational report and the availability of every CVE patch: yes, and by a wider margin than most of what this series examines. Is it an open institution in the Hayekian sense — a system in which dispersed knowledge, including knowledge held outside the originating firm and outside the country, can be transmitted into the artefact — partially, on the upstream axis, and not on the governance axis.
Checkable questions, all cheap.
(i) Over the next twelve months, does the ratio of mainline inclusion to vendor-original commits in OLK-6.6 rise or fall? Both numbers are mechanically countable from the mirror, which is precisely the value of publishing one. A falling ratio within the same volume is not a scandal; it is a drift, and it is measurable.
(ii) Does the GitHub mirror ever carry an issue — even one? A single issue opened and answered on openeuler-mirror would prove the mirror is a channel the project reads. Its permanent emptiness says the mirror is a publication.
(iii) Would the AtomGit conversation be readable from outside China under the same terms as the code? Publication without discussion is a fair arrangement for a distribution and an insufficient one for a root community, and the gap costs the project the one thing it says it wants globally: contributors who arrive from a search result rather than from a partnership.
(iv) Does AtomGit mirror openeuler outward for anyone who never joins the community? The answer is the difference between a sovereign forge and a localised one, and the two are entirely different institutional objects.
📡 WeChat Monitor — A Native-OS Critique That Found the Forklift and Stopped Before the Garage
The 2026-11-01 piece from 早安胡同 (Good Morning Hutong) — “喊着自主可控的国产系统,怎么最后都活成了Windows的样子?” (“Domestic systems that shout ‘self-reliant’, how did they all end up looking like Windows?”) — is the most institutionally literate thing in this week’s manual feed, and it deserves to be read as an argument rather than as a complaint. The claim is made in the author’s own terms, so it is quoted rather than paraphrased:
“自主研发不等于界面不能像 Windows” — self-reliant R&D does not mean the interface cannot look like Windows. Kylin and UnionTech UOS are largely Linux derivatives with their own DDE and UKUI desktop environments; the security kernel, permission control, 国密 (national cryptographic algorithm) hardening and log auditing “are real modifications”.
“兼容层是过渡期救命方案,不该是永久核心卖点” — the compatibility layer is a lifesaver for the transition period, but it should not be the permanent core selling point. Kylin’s KWRE and UnionTech’s Windows compatibility engine are, in the author’s description, “a translation layer that translates Windows software instructions into something Linux can understand”, so CAD, legacy finance tools and industry-specific packages can run without re-adaptation.
“系统本身变成了一个’Windows 软件启动器’” — the system itself has become a Windows software launcher. Native domestic software development is slowed down by the compatibility solution, because if the compatibility layer can run the old software, many units have no incentive to push developers toward native adaptation.
And the sentence that closes the piece as an institutional thesis rather than a complaint: “兼容层是过河的桥,不能当成一辈子住的房子” — the compatibility layer is a bridge across the river, it cannot become the house you live in for the rest of your life.
One correction to the piece, offered as evidence rather than as disagreement. The author treats Huawei’s HarmonyOS PC as the counter-example — a domestic system that did not copy Windows’ interaction model, going for distributed cross-device flow instead. The correction is that openEuler’s own DevStation image ships GNOME 49 with Dash to Panel and Arc Menu preinstalled, precisely “for a slightly more familiar layout”, and offers the Deepin Desktop Environment, UKUI (as seen in Ubuntu Kylin), MATE, Xfce and Kiran as alternatives. Even the project with the strongest claim to being a root community is, at desktop level, assembling other people’s desktops. That does not weaken the author’s argument. It strengthens it, and it locates the reason: the desktop layer has no institution behind it on any side of the industry. The foundation’s lifecycle judgement covers the kernel and the release; nobody’s lifecycle judgement covers the interaction paradigm.
Institutional reading. The piece is an unusually clean statement of what this series has called forklift acquisition — a switch driven by procurement with the working system preserved underneath. What it gets right is that the compatibility layer is not a technical choice but an incentive structure: the moment a legacy binary runs, the buyer’s marginal benefit from funding a native port goes to zero, and the policy that mandated the switch stops producing pressure for the thing it was supposed to produce. What it stops just short of is the next question — why is 信创 (the domestic-IT substitution programme) buying an operating system rather than buying maintainers? Every well-functioning ecosystem this series has examined is maintained by people whose salaries, tenure or reputation depend on the maintenance, and in a procurement-driven market the entity with the incentive to maintain is distributed across the user base, which is precisely the entity with no incentive at all. That is not a moral failure of the OSVs. It is what a metric-driven substitution programme does to the thing it substitutes for.
Deduplication note: this account is not one of the sixteen monitored official accounts, and it is appearing here for the first time (the manual feed entry dated 2026-11-04 covered a system-selection article, not this argument). It is used in this section only for the native-adaptation framework connecting to the openEuler material above; the compatibility-layer mechanics were covered by an earlier briefing and are not re-reported here as new.
🔍 Commentary — Two Technologies, One File
First: the licensing question has been settled for a decade and the governance question has not been settled anywhere. GNU GPLv2 (with the syscall exception) gives the world the right to take openEuler’s kernel and do anything with it. It has given that right for forty years. What GPLv2 cannot do — and has never claimed to do — is create an organisation that accepts a patch from a stranger in Berlin. Where that organisation exists (the mainline, Debian, Fedora), a project is a locus of decision; where it does not, a project is a distributed artefact. Both can be excellent engineering, and the difference between them is invisible in the licence file and obvious in an issue counter.
Second: the AtomGit migration is a fact about the stack, not about this repository. openEuler’s kernel left Gitee for AtomGit; Gitee’s landing page now forwards you, in two languages, and its 1.25-million-commit counter stops where the migration happened. That is one project making a routine hosting decision — the kind any project makes. What it also shows is the completion of an industrial stack under Chinese control: code hosting = AtomGit, package mirrors = MirrorZ, developer tooling = community editions, compliance auditing = an administrative standard. Each layer was built by a different kind of organisation — a vendor, universities, a community, a regulator — and each has a legitimate reason for existing. The institutional question is not whether the stack should exist; it is whether a project that lives in it can still be a place where strangers meet, because that property is not a feature of forges. It is a property of the rules the forge writes.
Third, and this is the one worth keeping: a mirror is a commitment to be readable, and readability is not governance — but it is not nothing either. The most industrially consequential open-source question of the next five years is not which country ships more open code. It is whether the world’s computing base has an audit trail that anyone can pull without permission. A nightly Git mirror, a signed CVE fix, a conflict noted in a commit message — those are the artefacts of that property, and they cost real money to maintain. OpenEuler produces them. The correct response from outside is neither applause nor suspicion, but the boring professional one: use the mirror, watch the ratio, and stop treating a download figure as though it were an institution.
Sources: The Register — openEuler brings China’s enterprise Linux ambitions to the desktop (2026-10-20) · Gitee — openeuler/kernel, migration notice to AtomGit, 1,245,880 commits on OLK-6.6 · AtomGit — openeuler/kernel (canonical: 590 stars, 676 forks, 1,849 issues, 1,486 PRs, 203 members; read 2026-11-05) · GitHub — openeuler-mirror/kernel (last push 2026-11-04T04:40:03Z; 59 stars, 41 forks, 0 issues, 0 PRs) · GitHub API — 400 commits on OLK-6.6, 2026-07-22 → 2026-11-04 (author and sign-off analysis) · openEuler — community operations report, September 2026 (2.9 万 developers, 2,199 members, 112 SIGs, 340.8K PRs; published 2026-10-07) · openEuler — Developer Day 2026 (2026-04-25): 16 million installs, 57.3% share, 2,100+ members · openEuler — Huawei Connect 2026 report: installs past 20 million (2026-09-20) · Huawei — 关于华为开源: 2,100+ member units, 25,000+ developers, 5.8M downloads, 16M installs, 2025 Graduation of openEuler and OpenHarmony · Beijing Municipal S&T Commission — 50+ open-source projects at the 2026 OpenAtom Ecosystem Conference (2026-07-01): 2,100+ units, 2.7万 developers, 16M installs, Graduation Nov 2025 · 199IT — Huawei: openEuler installs past 20 million, first in China’s server-OS market (Frost & Sullivan 2025 white paper: 4.48M new installs, 57.3%) · LWN — the current state of Linux architecture support (non-upstream kernel ports, 2025-11-18) · WeChat — 早安胡同, 喊着自主可控的国产系统,怎么最后都活成了Windows的样子?(2026-11-01) · IT之家 — 2026国产操作系统全景体验:八大系统谁主沉浮?(2026-05-28) · Sina Finance — 2026工业操作系统大会·钢铁行业工业操作系统创新应用论坛 (2026-11-04)