jschoubben
5e83ac2c22
Jochen: a normal application has 3-5 ADRs, maybe 10 for a large one, and we are at 65. Fair, and the cause is mine -- I recorded every FINDING as a decision rather than every fork in the road. Two merges, both cases where one decision had been split across many records because it was taken over several days rather than at once. 0019 absorbs ten records about how this repository works: what it is and that it is public, the folder flow, the two design layers, the issue front door, status in frontmatter, playbooks, the naming rule, the product name. Those were never ten decisions -- they were one, seen from ten angles as the repository took shape. 0016 absorbs the five about the lab: a node is a virtual machine, a router is scenery, a scenario declares the underlay, a scenario is a closed address space, and the two scenario classes. Same pattern -- one design, split by the order it was worked out in. The consolidated 0019 also raises the bar for what earns a record, since that is what produced 65: a record is warranted when there is a genuine fork -- a direction reversed, an alternative that will be proposed again, something contested. A finding is not a decision, and a bug is certainly not. Everything else belongs in the design document where the reasoning is actually read. The checker earned its place here. Deleting nine records left 13 dangling links across the repository and it named every one, including in AGENTS.md. Nothing was found by reading. Remaining clusters worth the same treatment: the host (8 records), delivery (5), modules (6), connectivity (4), substrate and control plane (4). That would be 52 down to roughly 30.
4.5 KiB
status, date, deciders, reconstructed, extends
| status | date | deciders | reconstructed | extends |
|---|---|---|---|---|
| accepted | 2026-08-28 | jochen | false | 0064-a-build-edge-is-a-third-kind.md |
65. The core library is the mesh's domain, and types ship with their modules
Context
ADR 0019 describes the shared library as "contracts shared across tiers: types, not behaviour" — a guard against what the current one became, which is a package holding too much code and, with it, everybody's dependencies.
The guard is aimed at the wrong thing. Types, not behaviour sounds safe and does not address the fault: a library everything depends on is a hub whether it holds types or code. The fan-in is what makes a change expensive, and types not behaviour leaves the fan-in exactly where it was.
ADR 0064 makes this measurable rather than a matter of taste: a module's cost is its inbound build edges, and a type hub has as many as a code hub.
Decision
Types ship with the module they belong to
A type is part of a module's contract, so it travels with the module. A consumer needing
inventory's types depends on inventory — a real edge, narrow and visible — instead of both
depending on a hub where the relationship cannot be seen.
This trades one wide edge for several narrow ones, and that is the improvement. Under ADR 0064 a change to one module's types now invalidates its actual consumers, rather than everything that touched the hub.
The core library holds the mesh's own domain, and that is the test
Not shared code. A drawer labelled shared is a drawer everything goes in, which is how the current one grew.
The core library holds what is true of the mesh regardless of which context you are in.
Research 011 already found what that is: a module, a node, and an assignment. Those three are the mesh's domain — every context speaks about them, and none of them belongs to one context.
The test, applied to a candidate: would this still mean the same thing in a context that had never heard of the one it came from? A node does. A pipeline stage does not — that is delivery's. A grant does not — that is provisioning's.
Domain-driven is the point rather than the label. The current library is what happens when the organising idea is who else might want this: the answer is always yes, so everything is admitted. Is this the mesh's domain has a defensible no.
Why this stays small on its own
A domain model changes when what the mesh is changes, which is rare. A shared-code drawer changes whenever anybody writes something reusable, which is constantly. So the core library inherits the property the graph is supposed to reveal — few changes, many dependants — rather than fighting for it.
Consequences
- ADR 0019's description of the shared library is superseded. Types, not behaviour is replaced by the mesh's domain, and its types move to the modules that own them. The rest of 0030 — the naming rule and the repository list — stands.
- A module now publishes its own contract, which it does not do today. That is real work and it is the same work as making a module a self-contained artifact, so it is not additional.
- The check is not the one that was proposed and is better. The build output contains no runtime code would have enforced types, not behaviour — a rule now withdrawn. What replaces it is inbound build edges, which is a measurement rather than a prohibition: the core library should have many, and anything else acquiring many is turning into a hub. That is visible while it happens rather than after.
- Fewer things will be shared, and some code will be written twice. That is the trade and it
should be said plainly: the current library exists because sharing felt free. Under this it has
a name, an owner and a visible edge, and two similar functions in two modules is often the
better answer —
how-we-build§8 already says three similar lines beat a premature abstraction. - Nothing here says how a module publishes its types, and the answer differs per language. That belongs with delivery.
References
- ADR 0019 — the description this replaces.
- ADR 0064 — what makes fan-in measurable.
- Research 011 — module, node, assignment.