OXYGENHUB
# e5b97ac399e7 oxygenhub / docs / diffusion.md
12.3 KiB · Markdown · last changed Diffusion reconciler: every 10 minutes the pin index is checked and mended 20 days ago renderedsource blame history raw
1
# Diffusion: upstream changes reach their consumers
2
3
Status: design, 2026-10-12. D1 is planned first; nothing is built yet.
4
"Diffusion" is a working name (changes diffuse from an upstream into the
5
repositories that pin it); Andy has not confirmed it.
6
7
## 1. The problem
8
9
R* API is a subrepo of Metal Gear, Metal Gear 2, Super Mario Land and
10
others. When it changes, each consumer should take the change, test it,
11
and propose it when it works, with no one opening proposals by hand.
12
Renovate and Dependabot do this for package manifests; Diffusion does it
13
for Oxygen pins, as part of Oxygen, and goes further than they can.
14
15
## 2. Why Oxygen can do better
16
17
- **The dependency graph is exact.** An Oxygen pin names a repository id
18
  and a state, and is a verifiable object (spec §7.8). The hub already
19
  reads every pin in a state to give jobs their subrepos. Nothing is
20
  scanned or guessed.
21
- **A pin is a set of changes, and independent changes commute.** A
22
  consumer can take an upstream state minus one change and whatever is
23
  built on it (§6). Git tools can only pin one commit: all or nothing.
24
- **Bumps never need rebasing.** The pin change commutes with the
25
  consumer's own work; only a consumer moving the same pin itself
26
  conflicts, as an ordinary replace conflict.
27
- **Everything is signed.** The bump, the upstream changes and the CI
28
  receipts are signed objects, so a consumer can check what it takes, even
29
  from another hub.
30
31
## 3. The approaches compared
32
33
| Approach | Strength | Weakness |
34
|---|---|---|
35
| Bump proposals (Renovate, Dependabot) | Consumers stay in charge; auditable; works across hubs | Reactive: breakage is found after upstream ships |
36
| Monorepo at head (Google) | Upstream cannot land what breaks a consumer | Takes control from consumers; upstream sees all code; one hub only |
37
| Upstream-driven migrations (large-scale changes, codemods) | Handles breaking changes, not just detects them | Needs tooling per change |
38
| Consumer-gated promotion (contract testing) | Prevents breakage | Needs consumers' tests to run upstream |
39
40
Diffusion is bump proposals as the base, with consumer checks on upstream
41
proposals (§7), partial pins (§6) and repair (§8) on top.
42
43
## 4. D1: follow and propose
44
45
### 4.1 The reverse pin index
46
47
When a push moves a repository's default bookmark, the hub records each
48
Oxygen pin in the new state: consumer, path, the upstream repository id
49
and the pinned heads (`diffusion_pins`). That answers "who consumes R*
50
API, at which state" at once. An upstream repository's page shows its
51
consumers on this hub ("Diffuses to"); the API gives
52
`GET /repos/{owner}/{name}/consumers`, and `GET /repos/{owner}/{name}/diffusion`
53
gives both directions with their pills (#153).
54
55
Keeping it right (oxygenhub#210). Each write records the state it indexed
56
(`diffusion_indexed`); a failed write is logged with the step that failed
57
and tried once more, and a state whose pins cannot be read leaves the
58
index as it was. Every ten minutes a reconciler compares each followed
59
bookmark with the state it was indexed at and re-indexes where they
60
differ; in the same pass, a consumer whose pin is not its upstream's
61
newest passing state, with no bump open or in draft, is caught up as if it
62
had just turned Diffusion on. A bump a person closed counts as a "no": it
63
is not opened again until the upstream moves. Repairs are logged
64
(`diffusion: reconciled: …`).
65
66
### 4.2 Turning it on
67
68
A consumer follows a subrepo when, in order of precedence:
69
70
1. its `.oxymodules` entry says so, beside the existing `url` and
71
   `bookmark` keys:
72
73
   ```toml
74
   [rstar-api]
75
   url = "https://oxygenhub.io/doomsdayone/rstar-api"
76
   bookmark = "main"
77
   follow = "green"   # off | green | any
78
   land = "propose"   # propose | auto
79
   ```
80
81
2. else its repository setting (Settings › Subrepos), so an
82
   administrator can turn it on without a commit;
83
3. else the hub's default (`[diffusion] follow` in hub.toml, `off`).
84
85
`green` waits for the upstream state's required checks to pass; `any`
86
follows every move. An upstream with no checks at all is **unverified**
87
(amber): it never turns green, so its moves bump green followers too, and
88
the proposal is labelled `upstream unverified` with a note saying only the
89
consumer's pipeline tested it (decided by Andy, 2026-10-13). A passing run
90
of the upstream later clears it.
91
92
**Landing by itself** (built 2026-10-14, oxygenhub#174; Andy chose
93
"opt-in, per repository" after rstar-api's first green/green): `land =
94
auto` in the entry, else the repository's Settings › Diffusion › Landing
95
(`PUT …/diffusion {"land": "auto"}`), off by default. The hub lands a bump
96
itself, as the `diffusion` account, when both halves are green: the exact
97
upstream state the bump pins passed its own run, and the consumer's
98
newest run on the bump bookmark passed on the bump as it is now. Never
99
for amber: an upstream with no checks always waits for a person. It lands
100
through the web's own path (`land_as`), signed by the hub's own key, so
101
the consumer's required checks, approvals and trust policy decide as for
102
anyone; a policy that does not let the hub's key move the bookmark keeps
103
the bump waiting, and the proposal's timeline says why, once. It is tried
104
when the bump's run passes, when the upstream's run passes under a bump
105
already made, and when the setting is turned to `auto`. The timeline
106
records "landed (Diffusion, both halves green)".
107
108
### 4.3 The bump
109
110
When an upstream's followed bookmark reaches a new state S (and, for
111
`green`, S's required checks pass), for each consumer that follows it:
112
113
- **Skip** when the consumer's pin already holds S, or when S does not
114
  hold everything the current pin does (upstream moved backwards or was
115
  rewritten). Consumers only move forward.
116
- **Build** one change on the consumer's default heads that replaces the
117
  pin's blob with a pin of S. The hub signs it with its delegate key, as
118
  it signs wiki edits, authored by the hub's `diffusion` account.
119
- **Place** it on the bookmark `diffusion/<path>`: the consumer's default
120
  heads plus the change. One bookmark per consumer and subrepo, moved in
121
  place when upstream moves again, so there is never a pile of bumps.
122
- **Propose** it, once: a proposal from `diffusion/<path>` to the default
123
  bookmark, labelled `dependencies`, kept up to date as the bookmark
124
  moves. Its CI runs on the projected state like any proposal's.
125
126
A trust policy that does not admit the hub's key still gets the
127
proposal; it is landed from a client, as today.
128
129
### 4.4 What the proposal says
130
131
- The upstream, the old pin and the new state.
132
- **The exact changelog:** every upstream change in S and not in the old
133
  pin, with its author, whether it is signed, and the issues it closes or
134
  refers to.
135
- Its checks, as for any proposal. A red bump stays open, with the
136
  failing check; D2 then says which upstream change broke it.
137
138
### 4.5 Seeing it (#150)
139
140
A repository's **Diffusion** tab lists who pins it (Used by) and what it
141
pins (Follows), each with a status Andy chose on 2026-10-13: **green**
142
(the upstream was checked and the consumer's run on the bump passed),
143
**amber** (the run passed but the upstream has no checks), **red** (the run
144
failed), **grey** (running), **up to date**. The overview shows the same in
145
brief; Admin › Diffusion shows the whole hub.
146
147
It is drawn as a two-half pill, like a capsule (Andy, 2026-10-13): the
148
left half is the upstream (**checked**, **untested**, **failing**,
149
**testing**), the right half the consumer (**passes**, **fails**,
150
**testing**, **waiting**, **up to date**), each coloured on its own, so a
151
game that passes on an untested upstream reads [untested │ passes] rather
152
than a single amber that hides the pass.
153
154
## 5. Across hubs
155
156
Each hub only ever writes to its own repositories. What one hub tells
157
another is a hint; the consumer's hub fetches and verifies the facts.
158
159
- **Polling always works.** The consumer's hub has the upstream URL in
160
  `.oxymodules`; it asks for the followed bookmark every few minutes. An
161
  upstream kept in a Strata bucket with no hub is polled by reading its
162
  bookmark record.
163
- **Subscribing makes it prompt.** The consumer's hub asks the upstream's
164
  hub to notify it when the bookmark moves. Notes are signed with the
165
  upstream hub's key and carry only "repository *id* moved to state *S*";
166
  the consumer fetches and verifies. A forged note causes a fetch, nothing
167
  more; a lost one is caught by the next poll.
168
- **Private upstreams** are read with a read-only credential the
169
  consumer's owner stores as a secret on the consumer's hub.
170
- **"Green" across hubs** is evidence the consumer checks: a Flow receipt
171
  for S, signed by a runner the upstream hub vouches for. The consumer's
172
  policy names the upstream hub keys it accepts, and may also require
173
  every upstream change to be signed by the upstream's maintainers, which
174
  holds even if the upstream hub is compromised.
175
- **Consumer checks across hubs (§7)** are opt-in on both sides; the
176
  consumer's hub runs them on its own runners and returns only a signed
177
  summary, or nothing.
178
179
## 6. D2: partial pins and the culprit
180
181
A pin is a state: any set of changes closed under dependencies. So a
182
consumer can pin "upstream S, minus change X and everything built on X":
183
its heads are the maximal changes of `closure(S) − ({X} ∪ dependents(X))`,
184
well-formed by construction.
185
186
When a bump of N upstream changes fails:
187
188
1. Flow searches for the culprit by testing subsets of the N changes (a
189
   delta-debugging search over independent changes, bounded in runs).
190
2. The bump proposal then pins everything except the culprit and its
191
   dependents, and says which change is held back and why, with the
192
   failure.
193
3. The consumer gets the rest at once; the held-back change comes with a
194
   later bump, once upstream fixes it.
195
196
`oxy submodule` and checkout take such pins as they are (a pin is heads);
197
`oxy submodule status` names the changes a partial pin leaves out.
198
199
## 7. D3: consumers test upstream proposals
200
201
A proposal on R* API can run each consumer's pipeline with its pin swapped
202
for the proposal's projected state, committing nothing to the consumer
203
(the claim already carries subrepo pins; the run overrides one). Each
204
consumer's result appears on the upstream proposal as a check,
205
`consumers/<owner>/<name>`, with a summary: "Metal Gear ✔, Super Mario
206
Land ✘ (test X)". The person changing the API sees what breaks before it
207
lands, which is the monorepo's main strength without taking consumers'
208
control. An upstream may make consumer checks required.
209
210
## 8. D4: repair
211
212
- **Migrations with the change.** An upstream change that alters the API
213
  may carry a migration (a script, or instructions for an agent) under
214
  `.diffusion/migrations/`. A bump spanning it runs the migration in the
215
  consumer and adds the result as a separate change on the bump proposal.
216
- **An agent's fix.** When a bump is red, a scoped agent key (an agent
217
  home with `changes` scope on that consumer, never `refs`) may add a fix
218
  as a separate change to the proposal, for the consumer's owner to
219
  review. The bump and the fix stay separate changes, so either can be
220
  dropped.
221
222
## 9. D5: reach
223
224
- **Auto-land** (`land = "auto"`): a green bump lands itself when the
225
  consumer's trust policy and required checks allow it, and every
226
  upstream change is signed by keys the policy trusts.
227
- **Chains:** landing a bump moves the consumer's default bookmark, which
228
  may be followed in turn.
229
- **Grouping and quiet hours:** several upstreams' bumps in one proposal;
230
  no bumps outside set hours.
231
- **Cross-hub** polling and subscriptions (§5).
232
- **Channels:** with Cylinder, a consumer follows `rstar-api:stable`, a
233
  signed channel whose moves are gated by consumer checks (§7).
234
235
## 10. Order
236
237
| Step | What |
238
|---|---|
239
| D1 | The reverse pin index and Used by; turning it on; the signed bump on `diffusion/<path>`, moved in place; the proposal with the changelog |
240
| D2 | Partial pins; the culprit search |
241
| D3 | Consumer checks on upstream proposals |
242
| D4 | Migrations and agent fixes |
243
| D5 | Auto-land, chains, grouping, cross-hub, channels |
244
245
D2 is what no other tool can do; D3 is worth the most.
246
247
## 11. Open questions
248
249
- **The name:** Diffusion, or plainer ("Follow")?
250
- **Who may turn it on** for a consumer: its owner and administrators, or
251
  also the upstream's maintainers (by offering, which the consumer
252
  accepts)?
253
- **Bot identity:** a `diffusion` account per hub, or the upstream
254
  pusher's name with the hub as delegate?
255
- **Cost limits** for the culprit search and consumer checks: runs per
256
  bump, per upstream proposal, per day.