[coffeemuse]

Fixes #202.

What changed

The mode field's tooltip named the terminal from the profile (TerminalType.For). b3270 names it by its own rules (create_3270_termtype), so a colour model 4 or 5 showed 3279-4-E while the host was told IBM-3278-4-E, and a custom size showed 3279-2-E instead of IBM-DYNAMIC.

• Backend: IndicationParser reads b3270's terminal-name indication. B3270Session.TerminalName keeps the latest report, and a disconnect leaves it alone.
• Core: IEmulatorSession.TerminalName is null until the engine starts. TerminalType.For is now only the -model argument, and its comment says so.
• App: StatusFormatter.ModeTip shows the name as reported, IBM- prefix included, the way x3270's About box does. Before the first connection attempt the tooltip says only "Not connected".
• Docs: the user guide's status-bar passage names the terminal. Its mono row now says a mono profile tells the host "the terminal has no colour", because colour models 4 and 5 and every TN3270E connection also say 3278. The backend and Core CLAUDE.md notes and the CHANGELOG are updated, and the guide HTML is regenerated.

TN3270E

terminal-name and Query TerminalName both report termtype, the TELNET TERMINAL-TYPE name. Over TN3270E, b3270's DEVICE-TYPE request always asks for the 3278 form (tn3270e_request). Neither that request nor the host's DEVICE-TYPE IS reply (connected_type) is reported anywhere. The tooltip shows the reported name on every connection, as x3270 does, rather than LizTerm copying the RFC's 3278 rule. So a colour model 2 or 3 over TN3270E shows IBM-3279-2-E, and the user guide explains why. Checked in the 4.5ga6 source and in 4.6pre1.

Testing

• Parser: Parses_terminal_name.
• Session: Terminal_name_is_the_engines_latest_report_and_outlives_a_disconnect.
• Replays: IBM-3279-2-E from the TN3270E ibmlink-help.jsonl, and IBM-DYNAMIC from oversize-100x50.jsonl.
• Engine: EngineTerminalNameTests runs the bundled engine for colour models 2, 4 and 5, mono model 3 and a custom size. It checks session.TerminalName against Query TerminalName. This is the first real-engine confirmation of the model 4 and 5 names.
• App: StatusFormatterTests and SessionViewModelTests are updated, and FakeEmulatorSession has a settable TerminalName.
• dotnet test LizTerm.slnx: 0 failures. dotnet build --no-incremental: 0 warnings.
• Hands-on (macOS, against MVS/CE over TN3270): colour models 2, 4 and 5, mono model 3 and a 100x50 custom size each showed the expected name in the tooltip: IBM-3279-2-E, IBM-3278-4-E, IBM-3278-5-E, IBM-3278-3-E and IBM-DYNAMIC. MVS/CE served a screen at 100x50 to IBM-DYNAMIC. TN3270E wasn't checked live, because there's no TN3270E test host.

🤖 Generated with Claude Code

[coffeemuse]

High review: fixes in e3cead9

The review raised 8 findings. 5 are fixed, 2 skipped and 1 was wrong.

Fixed

1. TN3270E wording. The user guide, the backend CLAUDE.md note and the TerminalNameIndication comment said TN3270E always asks for a 3278. tn3270e_request uses create_3270_termtype(true), which is a 3278 even for a colour profile but still IBM-DYNAMIC for a custom size. All three now say so. (This PR's description has the same slip in its TN3270E section; this comment corrects it.)
2. The name now dies with its engine. It's cleared in OnProcessEnded and TearDown, as _tlsOptions is, and in OnProcessEnded before the Disconnected state is raised. That keeps IEmulatorSession's promise that a caller can read it on ConnectionChanged. A replacement engine that reports no name can no longer show its predecessor's. New test: A_dead_processs_terminal_name_does_not_survive_into_its_replacement. The two replay tests now read the name while the engine is running: the ibmlink one when the TN3270E session comes up, the way the App reads it, and the oversize one before the engine exits.
3. A report with no text keeps the last name. New test: A_terminal_name_report_without_text_keeps_the_last_name.
4. TerminalType's summary no longer names b3270 in Core. Root CLAUDE.md says Core "never mentions Avalonia, b3270 or mvsMF".
5. ModeTip builds the connection sentence once.

Skipped

• The tooltip refreshes only on ConnectionChanged. b3270 reports a new name before the connection change that follows it (net_connect sets termtype before any change_cstate in 4.5ga6 host.c), and nothing LizTerm does changes the name while connected. A separate change event can come later if a backend ever needs it.

Not a problem

• The name is read across threads without a barrier. It's the same pattern as Tls. The App marshals ConnectionChanged to the UI thread, and that hand-off orders the read after the reader thread's write.

dotnet test LizTerm.slnx: 0 failures. dotnet build --no-incremental: 0 warnings. The guide HTML is regenerated.