[vadi2]

Brief overview of PR changes/additions

• Route the tracked-link maintenance through the console model rather than the view. The main console's model outlives the TConsole built on it and keeps taking lines meanwhile, so stopping at mpConsole gave up adjusting line numbers while the buffer was still being trimmed - concealed links then pointed at the wrong lines and overwrote text on reveal, the very symptom clearWindow() leaves OSC 8 links pointing at lines that no longer exist Mudlet/Mudlet#10122 reports.
• Check that the buffer actually is the one the manager tracks, on both paths. TBuffer(host, console) is a legal way to build a scratch buffer, and its constructor calls clear() - which reached through the back-pointer and dropped every tracked link on the live console. Every model also builds its buffer before a view attaches, so the model path needs the same identity check.
• Drop a comment on test_aWholelineConcealmentMovesTheLinksBelowItOnlyOnce claiming a click is the only thing that conceals without a delay. concealLink() conceals unconditionally and the same test calls it four lines below.
• Tests: a link on the first removed line (the >= lower bound had nothing covering it), the queued screen-reader announcement being cancelled by a clear, a delete with no view attached, another buffer's construction and deletion leaving the tracked links alone, and a miniconsole maintaining its own links off its own buffer. friend class TrackedLinkTrimTest reads the line numbers and the pending announcement count, which are otherwise only observable through a screen reader.

Motivation for adding to Mudlet

The fix in this PR is right, but it addresses the manager through the view. Two states break that: a model with no view yet or no longer, and a second buffer that happens to hold a console pointer. Each was reproduced with a failing test before the change.

Other info (issues closed, discussion etc)

• TBuffer::shrinkBuffer() carries the identical if (mpConsole) gate, but it predates this PR (it came in with fix: hidden links still reveal once older text has scrolled away Mudlet/Mudlet#10119), so it is deliberately left alone here rather than growing the scope. Filed as A buffer trim corrupts tracked OSC 8 links when no view is attached Mudlet/Mudlet#10315 with a reproduction: a concealed link's commands get swept and its line number never moves.
• test_aWholelineConcealmentMovesTheLinksBelowItOnlyOnce passes against pre-PR code, so it is a double-adjustment guard rather than proof of this PR's fix - worth knowing given the commit message describes the manual loop's removal as covered by it.
• Two mutation survivors were judged not worth a test: stopping mpAnnouncementTimer in clear() and calling stopTimerIfNotNeeded() in adjustLineNumbers() are both timer hygiene with no observable behaviour to assert on.

Test case: cmake --build --preset linux-debug-nosan && QT_QPA_PLATFORM=offscreen ctest --preset linux-debug-nosan -R TrackedLinkTrimTest. Manually: open a game with OSC 8 links, conceal one, then have the game clear the window or push enough lines to trim the buffer - the concealed link must not reappear over unrelated text.

Assisted-by: Claude:claude-opus-5

[jaypatrickhoward]

Merged, thanks - this is on Mudlet#10179 now, along with development merged in.

Both fixes check out. I reverted each in turn and confirmed the tests bite: test_deletingALineMovesTrackedLinksWithNoViewAttached fails without the model path, and test_anotherBuffersLifecycleLeavesTheTrackedLinksAlone fails without the identity checks. The three additions that pass against unfixed code have teeth too - the miniconsole case fails if the mpConsole branch is dropped, test_clearingTheWindowDropsAQueuedHiddenAnnouncement fails if the announcement reset is reverted, and the first-removed-line assertion fails if adjustLineNumbers()'s lower bound is loosened to >. That last pair covers gaps we left untested, which is the more useful half. 164/164 on the merged result.

You were also right about the comment on test_aWholelineConcealmentMovesTheLinksBelowItOnlyOnce - the test calls concealLink() five lines above the click, and concealLink() only bails on untracked-or-already-concealed, so what we wrote there was wrong.

One leftover: registerHyperlink() at TBuffer.cpp:3563 is the last mpConsole gate on the manager in that file. Different consequence - the link is never registered rather than corrupted - so it may be fine as is. Noted on Mudlet#10315 rather than here, since that is where shrinkBuffer() is being tracked.

Assisted-by: Claude:claude-opus-5