Commit `8839e2cc95`

8839e2cc95d19fe316869d48d4f96dc3bd9b4e8d

parent: 34360906d0

Verified · cmc ci/build: success ci/test: success ci/vuln: success

cmc <hello@cleberg.net> · 2026-09-04 16:01 UTC

CHANGELOG: v1.12.0

Layout: unified · split

CHANGELOG.org +56

		@@ -4,6 +4,62 @@ Versioning follows semver from v0.1.0. Database migrations run
4	4	automatically on daemon start; upgrade notes appear per release when
5	5	anything beyond "replace the binary and restart" is needed.
6	6
	7	* v1.12.0 — 2026-09-04
	8
	9	The store and the push path. Dashboard queries walk an index instead of
	10	sorting the table, concurrent writers wait instead of failing, expired
	11	rows are swept, and a large first push no longer forks a process per
	12	commit.
	13
	14	- Every dashboard list scanned its table and sorted the result to take
	15	fifty rows. Reachability is correlated subqueries and cannot be
	16	indexed; the index supplies the =ORDER BY= instead, so the walk stops
	17	at =LIMIT=. On 20k issues and 20k merge requests: open issues 11.8ms
	18	to 0.8ms, open merge requests 12.1ms to 0.8ms, the review queue 15.5ms
	19	to 0.3ms, assigned issues 10.6ms to 0.04ms. The last of those also
	20	drives from =issue_assignees= rather than testing =EXISTS= against
	21	every issue. #137
	22	- Concurrent writers serialise rather than failing. Every transaction in
	23	the store writes, and a deferred one takes the write lock at its first
	24	write, by which point another may hold it; SQLite answers
	25	=SQLITE_BUSY= and does not run the busy handler for that case, so
	26	=busy_timeout= could not help. Measured with eight concurrent
	27	read-then-write transactions, 44% of them failed. Beginning immediate
	28	takes the lock up front, where the timeout applies. #121
	29	- =repo settings= updates no longer overwrite each other.
	30	=settings_json= is one blob, and every caller read it off a repository
	31	loaded earlier, changed a field and wrote it all back; two admins at
	32	once and the later write put back what it had read for the other's
	33	field. The read and the write happen together now, under one
	34	immediate transaction. #123
	35	- Expired sessions and tokens are swept, and =[retention]= caps how long
	36	the audit log, the activity feed, webhook deliveries and the mail
	37	queue keep a row. Unset means forever, which is what every existing
	38	instance gets. #122
	39	- The audit entry records flag names without their values. Secrets never
	40	reached argv — they travel on stdin — but prose did: =--body <the
	41	whole issue>= was stored verbatim, in a table nothing pruned, for a
	42	repository that may be private. Identifiers are positional and
	43	survive. #122
	44	- pre-receive on a require-signed repository forked a =cat-file= per
	45	incoming commit and built one JSON message holding the whole push. A
	46	50k-commit first push forked 50k processes and held the history in
	47	memory twice. One =cat-file --batch= serves the push now, and the
	48	daemon verifies each commit as it arrives. #100
	49	- The payloads other surfaces decode are named types in =control=. httpd
	50	had its own copies of what the build, profile and dashboard commands
	51	emit, so a field added to a command was silently absent on the page.
	52	#126
	53
	54	Migration 0035 adds three indexes; it runs on daemon start.
	55
	56	Two changes in behaviour rather than configuration. A heavily contended
	57	write now blocks for up to =busy_timeout= (five seconds) instead of
	58	returning an error immediately — waiting is the point, but a caller that
	59	treated the error as normal will see a pause instead. And anything
	60	parsing =audit --json= for command arguments will find flag values gone;
	61	the flag names and the positional arguments are still there.
	62
7	63	* v1.11.0 — 2026-09-04
8	64
9	65	The web catches up with the rest of the forge: search across the

Commit 8839e2cc95

Commit `8839e2cc95`