[bitbonsai]

Problem

A single IRC message sent to multiple targets has the same msgid in each target's history. DeleteMessage receives a target and checks permission against it, but persistent-history lookup used only the msgid with LIMIT 1. It could select and delete a copy from another target, leaving the requested copy in place. In some paths, the wrong row could also trigger a permission failure for a valid request.

Deleting every row with that msgid would be worse: it would erase history from targets the delete did not name, without a corresponding REDACT there.

Change

Scope history lookup and deletion by both msgid and the casefolded sequence.target in SQLite, MySQL, and PostgreSQL. Only the requested target's row is affected; copies in other targets remain intact. This keeps the database mutation aligned with the target used for authorization and avoids cross-target data loss.

Regression test TestDeleteMsgidIsTargetScoped stores one msgid under two targets, verifies lookup selects the requested target, deletes that copy, and confirms the unrelated copy remains.

Checked in golang:1.27: go test -tags "i18n mysql postgresql sqlite" ./... and go vet pass; gofmt is clean. MySQL and PostgreSQL were compile-tested, not run against live databases.

I am new to this codebase, so please flag it if I have misunderstood the target-scoping semantics.

[slingamn]

A single IRC message sent to multiple targets has the same msgid in each target's history.

I cannot reproduce this, e.g.

>>> privmsg #chat,#test :hi
@msgid=zbmw94cut24r5zrh9sisi36p2e;time=2026-10-05T15:30:04.343Z :c!~u@mzye9rsrvmrrc.irc PRIVMSG #chat :hi
@msgid=97fyebhh6s2zbj9z2cjc3rj7yn;time=2026-10-05T15:30:04.349Z :c!~u@mzye9rsrvmrrc.irc PRIVMSG #test :hi

which accords with my understanding of the implementation here:

ergo/irc/handlers.go

Lines 2469 to 2471 in 6e25291

	// each target gets distinct msgids
	splitMsg := utils.MakeMessage(message)
	dispatchMessageToTarget(client, clientOnlyTags, histType, msg.Command, targetString, splitMsg, rb)

[bitbonsai]

You're right, thanks for checking. I misread the multi-target path: MakeMessage runs per target, so each target gets its own msgid and its own history row, and the LIMIT 1 lookup can't hit another channel's copy. The premise of this PR doesn't hold, so I'm closing it.

The one place I found where a msgid does repeat across channels is NICK/QUIT events, since the same item is written to every channel the user is in (nickname.go, client.go in destroy). That seems too niche to be worth target-scoped deletes in three backends, but I'm happy to revisit if you think otherwise.

[slingamn]

Thanks, I opened #2464 to look into this