mesh-lab/replays/register.go

// Package replays is the lab's replay of every core incident (novox/hq to-be 45 §9, M9): a scripted
// replay of what happened, asserting the rule's outcome, run before every merge of a core repository —
// and proved, once, to fail on the commit before its fix and to pass on the fix.
//
// **Where a replay lives.** A replay of one component's logic is a test in that component's repository,
// written only with what the component had before its fix, so it can be laid over the older commit
// (the controller's replays_test.go). A replay of what the mesh runs rather than what it wrote — the bus
// server's release, the resolver the catalogue configures, under both C libraries — lives here, where a
// container runtime and a bus of a given release are at hand. Every one is in Register, with the issue,
// the fix, and how the prover runs it at a commit.
//
// **A new core issue resolves with its replay added here, or with a stated reason none is possible**
// (hq 00-META/checks/cycle.py holds the issue to it: `replay:` names an entry here, or `replay-none:`
// says why).
package replays

// Kind is how a replay is run at a commit.
type Kind string

const (
	// InRepository is a test in the fixed repository, laid over the commit and run there with go test.
	InRepository Kind = "test"
	// Bus is the bus replay here, run against a server of the release the catalogue pinned at the commit.
	Bus Kind = "bus"
	// Resolver is the resolver replay here, run with the catalogue's resolver configuration at the commit,
	// asked by a program under musl and one under glibc.
	Resolver Kind = "resolver"
	// Gate is a test of the merge gate itself: before its fix there was no check to fail, so the commit
	// before is held to fail by not having it.
	Gate Kind = "gate"
	// Liveness is the crash-loop replay here: a container that exits at start, raised on the runtime here,
	// judged by the node-engine at the commit, and the controller at the commit judging its gate from what
	// the engine said (novox/hq ADR 0240). Two repositories move together: Repository and With.
	Liveness Kind = "liveness"
)

// Also is another repository a replay runs at its own commits beside Repository's: before its fix and on
// it, as Repository is.
type Also struct {
	Repository string
	Fix        string
	// Before is the commit the replay runs this repository at before the fix, when that is not Fix's
	// first parent.
	Before string
}

// Replay is one incident, replayed.
type Replay struct {
	ID    string
	Issue int
	// What is the incident in a line, and Asserts the rule's outcome the replay holds.
	What, Asserts string
	Kind          Kind
	// Repository is where the fix is, and Fix the fix's commit there: the replay must fail on Fix's first
	// parent and pass on Fix.
	Repository string
	Fix        string
	// Before is the commit the replay must fail on, when that is not Fix's first parent: a fix still on its
	// branch is proved against the main it branched from. Set to "" once the fix is merged.
	Before string
	// Package and Test are where the replay is run: a package of Repository for InRepository and Gate,
	// of this module otherwise; Files are the files laid over the commit, from where the replay lives.
	Package string
	Test    string
	Files   []string
	// Home is where the replay's files live, when that is not the fix: the repository and ref.
	Home, HomeRef string
	// With is the other repositories the replay runs at their commits beside Repository's (Liveness).
	With []Also
	// Module is the directory of the Go module the replay runs in, within Repository, when that is not its
	// root: a catalogue module is a Go module of its own. Package is relative to it; Files are not.
	Module string
}

// Register is every replay, by incident.
var Register = []Replay{
	{ID: "R236", Issue: 236, Kind: Gate, Repository: "mesh-controller", Fix: "feat/merge-gate",
		Before:  "origin/main",
		What:    "a login manager's service passed the catalogue check and was refused whole by the node-engine",
		Asserts: "a manifest the node-engine refuses fails its pull request, naming the machine and the module",
		Package: "./cmd/mesh-controller", Test: "TestIssue236", Files: []string{"cmd/mesh-controller/merge_gate_test.go"}},
	{ID: "R262", Issue: 262, Kind: Resolver, Repository: "mesh-catalog", Fix: "20603b63e67323f2c725b242c0ea87281d4dee93",
		What:    "an Alpine container could not find a machine by its mesh name: NXDOMAIN for IPv6 is final to musl",
		Asserts: "every machine's name resolves through the catalogue's resolver under musl and under glibc",
		Package: ".", Test: "TestReplay262"},
	{ID: "R263", Issue: 263, Kind: InRepository, Repository: "mesh-controller", Fix: "6d620f77c3f3f967be953ff760823c6a51b7b8d3",
		What:    "a consumer's 26-character identity refused the resolver holder's whole declaration",
		Asserts: "a provider's machine composes whatever its consumers are called",
		Package: "./cmd/mesh-controller", Test: "TestReplay263", Files: []string{"cmd/mesh-controller/replays_test.go"},
		Home: "mesh-controller", HomeRef: "feat/replays"},
	{ID: "R266", Issue: 266, Kind: Bus, Repository: "mesh-catalog", Fix: "0275c2e",
		What:    "a merge on the bus was never handed to the controller: 2.10 skipped messages on a consumer with several filters",
		Asserts: "a consumer filtered like the controller's is handed every followed message under the mesh's traffic",
		Package: ".", Test: "TestReplay266"},
	{ID: "R273", Issue: 273, Kind: InRepository, Repository: "mesh-controller", Fix: "8bfaf1523ed2dddde80daa0994f1044e59db44b1",
		What:    "a rule for the resolver re-bound a machine's database consumers to empty databases elsewhere",
		Asserts: "a consumer beside its store stays bound to it while another machine holds the store's seat",
		Package: "./cmd/mesh-controller", Test: "TestReplay273", Files: []string{"cmd/mesh-controller/replays_test.go"},
		Home: "mesh-controller", HomeRef: "feat/replays"},
	// The controller's incidents of 2026-10-07, each a test laid over the merge before its fix. Home is
	// the controller's commit that added them; it stays reachable once that pull request merges.
	{ID: "R296", Issue: 296, Kind: InRepository, Repository: "mesh-controller", Fix: "3808634",
		What: "a plan's clock restarted at every save: a build queued for minutes read \"building for 1s\", " +
			"then 4s, then 9s, and was never LATE",
		Asserts: "a plan's line counts from when it entered its tier, and a plan standing still in its tier is not " +
			"saved again",
		Package: "./cmd/mesh-controller", Test: "TestReplay296", Files: []string{"cmd/mesh-controller/replays_test.go"},
		Home: "mesh-controller", HomeRef: "7597294ff86383c171b8bec3920b5472fbce00e9"},
	{ID: "R299", Issue: 299, Kind: InRepository, Repository: "mesh-controller", Fix: "c37d774",
		What:    "a seat's first verbs, all optional, made the self-check say its holder silent on every machine",
		Asserts: "a holder of a seat whose verbs are all optional is not said silent when it answers nothing",
		Package: "./cmd/mesh-controller", Test: "TestReplay299", Files: []string{"cmd/mesh-controller/replays_test.go"},
		Home: "mesh-controller", HomeRef: "7597294ff86383c171b8bec3920b5472fbce00e9"},
	{ID: "R300", Issue: 300, Kind: InRepository, Repository: "mesh-controller", Fix: "e7dca6c",
		What: "a merge that added a module said it changed nothing the mesh holds, and built nothing, so the " +
			"module was never registered",
		Asserts: "a merge adding a module asks the build seat for it, at the branch merged into, and opens no plan",
		Package: "./cmd/mesh-controller", Test: "TestReplay300", Files: []string{"cmd/mesh-controller/replays_test.go"},
		Home: "mesh-controller", HomeRef: "7597294ff86383c171b8bec3920b5472fbce00e9"},
	// A drill counted as a repair (2026-10-07): S15 read two deliberate drills, recorded through hand-act
	// record for want of a verb, as a cause repaired twice. In a file of its own, since replays_test.go holds
	// replays of later commits that do not build at the commit before this fix. Home is the controller's
	// commit that added it; it stays reachable once that pull request merges.
	{ID: "R292", Issue: 292, Kind: InRepository, Repository: "mesh-controller", Fix: "863ebd4",
		What: "two drills of ADR 0240, each with the operator's word, recorded through hand-act record --cause drill, " +
			"raised mesh.hand-acts.drill.healer-wanted",
		Asserts: "a drill recorded in the hand-act log wants no healer, however it was recorded",
		Package: "./cmd/mesh-controller", Test: "TestReplay292", Files: []string{"cmd/mesh-controller/replay292_test.go"},
		Home: "mesh-controller", HomeRef: "80f9d26e6d4c60ec7d137110b9d84c841072f04e"},
	// A seat's new verb deadlocked across two repositories (2026-10-07): the controller promising checks
	// refused the delivery seat's holder that did not serve it, and a holder serving it was refused by the
	// controller that did not promise it.
	{ID: "R298", Issue: 298, Kind: InRepository, Repository: "mesh-controller", Fix: "f6aea4b",
		What: "the delivery seat's new verb checks could land in neither repository first: each side's claim " +
			"check refused the other's holder",
		Asserts: "the delivery seat's holder holds it both before and after it serves checks",
		Package: "./internal/catalogue", Test: "TestReplay298", Files: []string{"internal/catalogue/replays_test.go"},
		Home: "mesh-controller", HomeRef: "80f9d26e6d4c60ec7d137110b9d84c841072f04e"},
	// The push a recorded build waits for, counted as a repair (2026-10-07): a test in the controller, in
	// the fix's own commit, recording two such pushes as the seat's push records them.
	{ID: "R301", Issue: 301, Kind: InRepository, Repository: "mesh-controller", Fix: "e92a3fe",
		What: "the operator's pushes of new builds whose upgrade policy is record — the resolver, the network " +
			"managers, the packet filter — were counted as repairs, and S15 wanted a healer for each cause",
		Asserts: "a push that moves only recorded builds, recorded as the push records it, wants no healer " +
			"however often its cause is given",
		Package: "./cmd/mesh-controller", Test: "TestReplay301", Files: []string{"cmd/mesh-controller/replays_test.go"}},
	// The lab's Go replays never compiled before a lab merge (2026-10-07): the build agent ran a repository's
	// check in one toolchain. Before the fix there was no part to run, so the commit before is held to fail
	// by not having it.
	{ID: "R302", Issue: 302, Kind: Gate, Repository: "mesh-controller", Fix: "98ef7ba",
		What: "mesh-lab's replays register merged with NOT CHECKED HERE: its check ran in the TypeScript " +
			"toolchain, which holds no Go compiler",
		Asserts: "a repository in two languages has each declared part of its own check run in that part's " +
			"toolchain, and the layer passes only when every part does",
		Package: "./internal/builder", Test: "TestARepositoryInTwoLanguagesIsCheckedInBoth",
		Files: []string{"internal/builder/check_test.go"}},
	// A recheck left the old verdict standing (2026-10-07): the forge keeps the newest status of each context
	// on a head, and a recheck asks of the same head, so the old green let the pull request merge while the
	// fresh check ran. A test in mesh-delivery, which asks the forge's holder (core) to put the merge check
	// back to pending; the holder's own half is its TypeScript test.
	{ID: "R305", Issue: 305, Kind: InRepository, Repository: "mesh-catalog", Fix: "74f912e",
		What: "a pull request rechecked at 23:04 kept mesh/merge-gate and mesh/repo-check success from 16:59, " +
			"and the forge would have merged it while its fresh check ran",
		Asserts: "a recheck puts the head's merge check statuses back to pending before it asks the check, and " +
			"is refused whole when the forge cannot be told",
		Module: "modules/mesh-delivery", Package: "./cmd/mesh-delivery", Test: "TestARecheck",
		Files: []string{"modules/mesh-delivery/cmd/mesh-delivery/recheck_test.go",
			"modules/mesh-delivery/cmd/mesh-delivery/fakes_test.go"}},
	// A pull request ran its own copy of the repository's check (2026-10-08): the build agent read
	// merge-check.sh from the change's head, so a branch cut before the script, or one that deleted or
	// gutted it, merged with none of its tests run. Written with only what the agent had before the fix, so
	// it fails on the commit before by answering the warning. The fix is still on its branch: proved
	// against the main it branched from.
	{ID: "R310", Issue: 310, Kind: InRepository, Repository: "mesh-controller", Fix: "cb0327d",
		Before: "1a50d6e",
		What: "the media catalogue's Lidarr pull request, cut before its repository had a merge-check.sh, was " +
			"answered no repository check defined as a warning and merged with none of its tests run",
		Asserts: "where the base branch holds a merge-check.sh, its script judges the change's tree, and a head " +
			"without one is never answered the warning",
		Package: "./internal/builder", Test: "TestReplay310", Files: []string{"internal/builder/replay310_test.go"}},
	// Two order rules ordered one pair both ways (2026-10-08): the controller's member moved the build agent,
	// which builds the node-engine, and the node-engine goes before the controller, so the group
	// feat/a-machine-joins-through-the-tunnel was refused as a cycle and merged by hand. A test in the
	// controller, using only what orderOf had before its fix, laid over the commit before it.
	{ID: "R309", Issue: 309, Kind: InRepository, Repository: "mesh-controller", Fix: "c5a2edf",
		What: "a delivery group of the controller, the node-engine and the lab, each ready, was rejected: built by " +
			"put the controller's member first and engine before controller put the node-engine's first",
		Asserts: "rules that order one pair both ways are resolved by precedence — built by over engine before " +
			"controller — and the group is ordered, the controller's member first",
		Package: "./cmd/mesh-controller", Test: "TestReplay309", Files: []string{"cmd/mesh-controller/replay309_test.go"}},
	// An answer larger than one message of the bus (2026-10-08): the controller's conditions and status
	// outgrew the bus's max_payload once 689 stalled deliveries were open conditions, the client library
	// refused every reply inside the controller, and every caller timed out while calls said answered.
	{ID: "R314", Issue: 314, Kind: InRepository, Repository: "mesh-controller", Fix: "175b28e",
		What: "conditions and status answered three times what one message of the bus carries; the reply was " +
			"refused inside the controller and every caller waited 30 s and read that it did not answer",
		Asserts: "an answer larger than one message arrives whole to a caller that pages, and one that does not " +
			"is told so in a message that fits — never a timeout",
		Package: "./internal/link", Test: "TestReplay314", Files: []string{"internal/link/replay314_test.go"}},
	// One module's wait for a new login held every other module's walk (2026-10-08): the walk of the builds
	// waiting for a gate sent the laptop three moves in one send, the lighting module's account said relogin
	// needed and its daemon's unit failed in that account's manager, and the gate failed the send at its bound,
	// put the module back (taking its account group back) and stopped every walk behind it. A test in the
	// controller, reading the statements as the node-engine's JSON, so it is laid over the commit before.
	// Fix is the head of the fix: e3b5c22 alone raises the new condition kind without the plain words the
	// controller's suite requires since hq ADR 0253, which 51d7bbf gives it. Before is the main both stood on.
	{ID: "R318", Issue: 318, Kind: InRepository, Repository: "mesh-controller", Fix: "51d7bbf", Before: "6df30b6",
		What: "a send of three modules to the laptop failed its gate after ten minutes because one module's account " +
			"needed a new login; the module was put back and the walks of the other modules stopped behind it",
		Asserts: "a wait for a person's new login passes the gate with the wait carried and said as the module's " +
			"relogin-needed condition; nothing is put back, and the walk goes on to the next machine",
		Package: "./cmd/mesh-controller", Test: "TestReplay318", Files: []string{"cmd/mesh-controller/replay318_test.go"}},
	// An assignment answered as unknown while its module's build ran (2026-10-08): the merge adding `sensors`
	// asked for its build (issue 300), and `assign g14 sensors` a minute later said "no module of that name",
	// which read as a module nobody registered. A test in the controller, written with only what it had
	// before the fix, laid over the commit before. The fix is still on its branch: Fix is its pull request's
	// head, proved against the main it is rebased on.
	{ID: "R325", Issue: 325, Kind: InRepository, Repository: "mesh-controller", Fix: "8adb7f1", Before: "fe00fec",
		What: "a minute after the merge that added sensors asked for its build, assign g14 sensors answered no " +
			"module of that name, and the same call worked minutes later",
		Asserts: "an assignment made while its module's build runs says the build, is kept, and is made when the " +
			"build registers the module",
		Package: "./cmd/mesh-controller", Test: "TestReplay325", Files: []string{"cmd/mesh-controller/replay325_test.go"}},
	// A declared tool check never asked on a live machine (2026-10-08): the queue reached the bus through an
	// assertion only OverNATS met, and the NATS link the node-engine attaches had no Ask, so every tool check
	// read "no link to the bus is open" while the link was up. A test in the node-engine, written with only
	// what it had before the fix; Home is its commit that added it.
	{ID: "R331", Issue: 331, Kind: InRepository, Repository: "mesh-host", Fix: "824911d",
		What: "the sensors module's tool check on the laptop read no link to the bus is open for hours while the " +
			"node-engine's link was up and the same tool answered through the mesh MCP server",
		Asserts: "a tool check on the NATS link the node-engine attaches reaches that link, and is not refused as no link",
		Package: "./internal/link", Test: "TestReplay331", Files: []string{"internal/link/replay331_test.go"},
		Home: "mesh-host", HomeRef: "5684a4579d92"},
	// Not a core incident, and the first of its kind: a module's. The crash loop was found by a person
	// reading the agent server's log for another reason (issue 268); research 032 measured it, and ADR
	// 0240 makes the node-engine judge it and the gate fail it.
	{ID: "R-crashloop", Issue: 268, Kind: Liveness, Repository: "mesh-controller",
		Fix:  "9d15f3a",
		With: []Also{{Repository: "mesh-host", Fix: "03031a4"}},
		What: "the agent server crash-looped about a hundred times while the mesh read it applied, its tools served " +
			"and nothing raised: the gate judged a module by what the mesh saw from outside",
		Asserts: "a container whose program exits at start is said unhealthy by the node-engine, and its build fails " +
			"its gate on the first machine and is put back there",
		Package: ".", Test: "TestReplayCrashLoop", Files: []string{"cmd/mesh-controller/replays_test.go"}},
	// A module's again: the silent web application, which no liveness and no TCP check sees, raised by the
	// HTTP check its module declares (ADR 0240 Phase B).
	{ID: "R145", Issue: 145, Kind: Liveness, Repository: "mesh-controller",
		Fix:  "7f25666",
		With: []Also{{Repository: "mesh-host", Fix: "56e2ebe"}},
		What: "a web application accepted TCP and answered nothing for eleven hours while the mesh said its machine " +
			"healthy: its port was open and its program ran",
		Asserts: "a web server whose application never answers is said unhealthy by its declared HTTP check within two " +
			"looks, and the controller raises the module's condition on the second",
		Package: ".", Test: "TestReplaySilentWebApp", Files: []string{"cmd/mesh-controller/replays_test.go"}},
}

// Find is the replay of that id, or of that issue.
func Find(id string) (Replay, bool) {
	for _, r := range Register {
		if r.ID == id || r.ID == "R"+id {
			return r, true
		}
	}
	return Replay{}, false
}