1376 Commits

Include renames

Author	SHA1	Message	Date
mesh-tools	3b03c220ad	Check that the catalogue holds no module mesh-delivery but through the names its handover needs, and say how the seat passes (issue 1351)	2026-10-20 00:59:36 +00:00
mesh-tools	d16fa1e052	Hear the gate's transitions beside today's delivery module's in mesh-issues, and name the gate among the owners (issue 1350)	2026-10-20 00:58:01 +00:00
mesh-tools	6a39da5485	Owe the forge its statuses, views, notes and comments as keys, read the plan from plans as data, and call an empty check answer an error (issue 1349)	2026-10-20 00:52:42 +00:00
mesh-tools	beafa55d97	Ask the Approve through the gate's asks, hand merges to the bridge one member at a time, and move source.main once a group is merged (issue 1348)	2026-10-20 00:45:02 +00:00
mesh-tools	dd462a80e2	Keep the gate silent until it holds the seat, and carry today's state over once when it first does (issue 1347)	2026-10-20 00:29:19 +00:00
mesh-tools	a390bb40f8	Take a stray key out of the gate's reads (issue 1346)	2026-10-20 00:19:29 +00:00
mesh-tools	09f0b0d5ab	Move the delivery module to the gate: mesh-delivery-gate claims the seat with its own buckets and reads (issue 1346)	2026-10-20 00:19:19 +00:00
jochen	b9104454f9	Read the grant file on keycloak's moved path, so a consumer not moved is still made (issue 1332) ... mesh-mail-web, grafana and car-hunter are not moved, so keycloak wanted nothing for them: it never made mesh_novox_mail_web, and held mesh_ace_grafana and mesh_ace_carhunt for retirement. Keycloak's moved spec now asks for the grant file's consumers too (WithGrants: true), and the moved harness reads MESH_RECEIVES exactly as the grant path does and as postgres's moved harness already did.	2026-10-20 00:30:45 +02:00
jochen	73a10fafc7	Pull keycloak's moved-path harness into movedHarness, and test it reads the grant file (issue 1332) ... Adds the failing unit tests for keycloak's provider on the moved path: a grant-file consumer not moved should still be made and never held for retirement, and the moved harness should read MESH_RECEIVES as the grant path does. All three fail on an assertion against today's code, which wants nothing from the grant file on this path.	2026-10-20 00:30:10 +02:00
jochen	9f3a8ee391	Follow real links instead of building them, and check the CSP on every answer (issue 1268)	2026-10-19 23:18:53 +02:00
jochen	cad6194447	Keep the HTTP endpoint's own state, so status and health stop claiming to serve once it has failed (issue 1283)	2026-10-19 23:18:51 +02:00
jochen	3fd1b47fe7	Bounce once through this origin before sign-in, so a Strict cookie is not lost crossing from Keycloak or another site (issue 1263)	2026-10-19 23:18:48 +02:00
jochen	5b57422523	Report an unknown ask through a small interface, not *testing.T directly (issue 1285) ... A failed subtest fails its parent, so TestHarnessFailsTheTestWhenTheBundleAsksSomethingUnknown could never pass the way it was written (try 2's finding). The player now reports through askFailer (Helper, Errorf); every other test keeps passing its own *testing.T, which satisfies it already. That one test gives a recorder instead, and asserts it holds exactly one message naming seat:mail-store.accounts, and that mail-webclient.status's own answer shows the ask got an error back. Checked: no other integration test uses the same deliberate-subtest-failure pattern (grepped t.Run).	2026-10-19 22:53:18 +02:00
jochen	523d7846cd	Fix mesh-mail-web's manifest: operators is type list, the listen word names its port number (issue 1262) ... The controller refused the manifest: config.operators declared type "array" (not one of the controller's five types, novox/hq ADR 0348 §4 — the word is "list"), and the bundle's env word MESH_MAIL_WEB_LISTEN said ${port:web} (a bundle's word takes the port NUMBER a listens entry gives, novox/hq ADR 0192; mesh-controller internal/catalogue/port_into.go portsFilledInto matches it against listens[].Port, never an endpoint's label). Fixed to type "list" and ${port:8099} (the listens entry's own port). main.go and the integration harness already read/write a plain port number, so neither needed a change. manifest_test.go now asserts both: config.operators.Type == "list", and the listen word equals ${port:<the listens entry's port>}; confirmed each independently fails against the old values before this fix.	2026-10-19 22:47:56 +02:00
jochen	502cda34cb	Add design 59 milestone 1's acceptance tests on the runtime player (issue 1268) ... Four scenarios, entered only through HTTP with Host mail.<node>.internal as a browser would, against the built mesh-mail-web, a stand-in identity provider (auth_test.go's own fakeIssuer, reused) and a real account bucket holding granted account a and ungranted account c: sign-in redirects, 403s and the session cookie's attributes; the 404 for another host; browsing and a send followed to sent, only the five mail verbs asked, account c never shown, paging by date, no database or extra writable directory; a mail's script and remote image shown as text, an attachment downloaded with nosniff, and the CSP on every answer. Build tag integration; compiles under go vet -tags integration ./...; never run here.	2026-10-19 22:42:27 +02:00
jochen	c996121e9d	Add the runtime player and stand-in mail-store for a bundle's integration test (issue 1285) ... Plays the SDK's own runtime protocol (initialize, tools/list, tools/call, the bundle's mesh/ask and mesh/state.<verb>) against mesh-mail-web's real built binary, from stdio.go and state.go's own source. The account bucket is a real JetStream key-value bucket (mesh-mail_accounts, named as mesh-mail's own acceptance_integration_test.go names a module's state); its fixture (accountBucketValue/accountBucketFolder) is accounts.go's own, itself copied from mesh-mail's mirror.go AccountValue (issue 1255, 1264). Build tag integration; compiles under go vet -tags integration ./...; never run here (go test -tags integration is this batch's hard rule) — the lead's try runs it.	2026-10-19 22:38:36 +02:00
jochen	f863e852c9	Add the write page and a send's state, behind issue 1284's form token (issue 1267) ... Wires 1284's per-session token into the real write form and its real POST /send; adds tests for 1284/AC1-AC4 against that real page and handler, beside this issue's own AC1-AC4.	2026-10-19 22:29:41 +02:00
jochen	58b337770d	Show a mail as escaped text and serve its attachments as downloads (issue 1266)	2026-10-19 22:26:47 +02:00
jochen	ff1886ac62	Add the search page, by mail-store.search's own argument names (issue 1265)	2026-10-19 22:24:52 +02:00
jochen	28fdbbd346	Show every granted account, its state and its folders' counts (issue 1264) ... Accounts bucket fixture (accountBucketValue/accountBucketFolder) copied from mesh-mail's own modules/mesh-mail/cmd/mesh-mail/mirror.go AccountValue/AccountFolderValue (issue 1255); mesh-mail-web is its own Go module, so this is a copy, not an import.	2026-10-19 22:22:52 +02:00
jochen	3ee80c98ab	Bind every changing form's token to its session (issue 1284)	2026-10-19 22:18:03 +02:00
jochen	8e0ef84060	Sign in through Keycloak: code flow with PKCE, verified ID token, 12-hour session (issue 1263) ... Library choice: golang.org/x/oauth2 for the code exchange, github.com/coreos/go-oidc/v3 for the ID token's verification (issuer, audience, expiry, signature from the issuer's JWKS).	2026-10-19 22:16:15 +02:00
jochen	68c141849f	Serve mesh-mail-web's HTTP endpoint: internal host check, CSP, open (issue 1283)	2026-10-19 22:09:49 +02:00
jochen	a9ad306693	Add the module mesh-mail-web, with its status verb (issue 1262)	2026-10-19 22:07:36 +02:00
jochen	a3e9985c15	Keep each account's newest mails in a bucket, bounded past 81 accounts (issue 1282)	2026-10-19 21:56:04 +02:00
jochen	08200bc178	Keep each account's state in a bucket (issue 1255)	2026-10-19 21:50:12 +02:00
jochen	a9ff48e200	Answer a send's state only to its caller (issue 1261)	2026-10-19 21:44:12 +02:00
jschoubben	a76910a0c0	Say only the merge tool's two out of date refusals a warning, and never step a severity down (issue 1275, ADR 0442's cases)	2026-10-19 20:51:58 +02:00
jschoubben	dc86203d29	Say a refusal for an out of date verdict a warning for ten minutes, urgent after (issue 1275, operator's amendment, AC6)	2026-10-19 20:51:58 +02:00
jschoubben	301cd2287f	Answer gitea_commit_statuses in the bus test's runtime, and say a held, refused member once (issue 1275, review round 1)	2026-10-19 20:38:48 +02:00
jschoubben	524e3405e5	Ask a group's Approve only when every member may merge, and withdraw the ask when a required status turns (issue 1275) ... An Approve given hours after its ask must not rest on a check made at the ask: a required status that turns not a success while the ask waits takes the ask back, and its late answer merges nothing.	2026-10-19 20:32:22 +02:00
jschoubben	6e6d1f3e47	Merge a group whole or not at all: re-read every required status first, stop at the first refusal, say a held or half merged group urgent (issue 1275, 1276) ... An approved group was merged member by member, each on its own, so a refusal of the first did not stop the second. A group now merges in one pass in its order, after every member's required statuses are read again at its head, and a held or half merged group stays raised. A refusal naming the member's own repository check as missing asks that check (issue 1276).	2026-10-19 20:32:15 +02:00
mesh-tools	eda55ceb9e	Merge pull request 'Give the forge bridge a merge commit's trees in one call, and each open pull request's head branch (design 47 R3c, issue 1234)' (#374) from feat/delivery-parts into main	2026-10-19 18:22:09 +00:00
jochen	5c1ff4bb15	mesh-mail: dedup-archive picks the same copy in both stores when a message-id also sits in a second non-archive folder (issue 1248, review round 1 finding 1)	2026-10-19 15:33:36 +02:00
jochen	280aeb2f47	Answer each gmail mail once when no folder is named, keeping archived mail (issue 1248)	2026-10-19 15:30:44 +02:00
jochen	920199862e	mesh-mail: group-by-sender and order-by sender compare addresses case-insensitively in Postgres too, and a sender's shown name and address are its newest mail's, deterministically (issue 1230, review finding 6)	2026-10-19 15:09:44 +02:00
jochen	8dd2a42448	mesh-mail: mail_search's order defaults to descending when left out, matching its own schema text, and AC1's no-filter test proves ordering, limit and grant isolation together (issue 1232, review findings 5 and 7)	2026-10-19 15:05:23 +02:00
jochen	d1a9056a1c	mesh-mail: mail_search gains words/words-any, a word found in subject or body, since subject-words and body-words together only ever meant and (issue 1229, review finding 3; also proves AC9's type-or-name independently and AC2+unread combined)	2026-10-19 15:02:40 +02:00
jochen	a3476a7bf2	mesh-mail: from-someone-i-wrote-to recognises gmail's Sent folder name too, and stays scoped to one account (issue 1231, review findings 1 and 2)	2026-10-19 14:57:26 +02:00
jochen	1c3158e098	mesh-mail: mail_search's address filter never treats % or _ as a wildcard, so a literal domain filter does not leak every mail in Postgres (issue 1228, review finding 4)	2026-10-19 14:52:36 +02:00
jschoubben	75957fc3c0	Hash the root tree from the root's entries, since the forge names the commit (issue 1234) ... The listing's top sha is the commit's, not the root tree's (seen live, novox/mesh-delivery at 7f61817), so gitea_commit_trees now computes the root tree as git does, from the entries without a slash. The query keys recursive and per_page stay in the allow-list: read flags only.	2026-10-19 14:52:29 +02:00
jschoubben	ea83746297	Give the forge bridge a merge commit's trees in one call, and a pull request's head branch (issue 1234) ... gitea_commit_trees reads the commit's recursive tree listing and answers the root tree and the tree of every directory holding a module.json; a cut listing fails loudly. gitea_open_pulls entries gain head. The query keys recursive and per_page join the allow-list, read flags only. mesh-delivery decodes open pulls with json.Unmarshal into a struct (cmd/mesh-delivery/ports.go:405-410), which ignores the added key.	2026-10-19 14:50:19 +02:00
jochen	b60ee706d4	mesh-mail: mail_search's tool makes the query optional and offers every filter and order-by (one schema for mail_search and mail-store.search), validated before the index is ever asked (issue 1232)	2026-10-19 14:46:58 +02:00
jochen	4a60deb4a8	mesh-mail: mail_search finds a thread's replies and mail from someone the operator wrote to, so a reply or a known correspondent's junk mail can be found (issue 1231)	2026-10-19 14:39:59 +02:00
jochen	fc10031d22	mesh-mail: mail_search orders by date, sender, subject or size and groups by sender, so the newest, the largest or who writes most can be answered (issue 1230) ... Query's From becomes a list, OR'd (1228/AC11's several domains), since 1230's order-by needs it to test the ascending-order criterion end to end; To and Cc stay single, no criterion asking for more.	2026-10-19 14:36:53 +02:00
jochen	2985122856	mesh-mail: mail_search's Query gains subject/body words, attachment type/name and read/unread, so a word in one field, an attachment or a flag can filter mails (issue 1229)	2026-10-19 14:29:15 +02:00
jochen	20342f2172	mesh-mail: mail_search's Query gains from, to, cc and a date range, so a sender, a recipient or a date can filter mails without query words (issue 1228)	2026-10-19 14:22:14 +02:00
mesh-tools	7bb588ca1b	Merge pull request 'mesh-mail sends an approved mail from its own sender process, and tells a failed send once (ADR 0436, ADR 0438; issues 1172, 1193-1199)' (#371) from feat/mail-sender into main	2026-10-19 11:36:51 +00:00
jochen	a2c0c58677	mesh-mail: a redelivery of a message whose send already finished (sent, failed, refused or not proven) is a repeat, terminated without recording, emitting or telling anything again (operator-approved change to ADR 0436 clause 10; issue 1196)	2026-10-19 13:29:31 +02:00
jschoubben	e601e88e95	Keep two values of each issue in the tracker's bucket, since the log holds them all (issue 1221)	2026-10-19 12:38:55 +02:00