[dieseld23]

Cleans up the compiler warning noise in the build and, in the process, fixes four genuine defects that were hiding in it.

Results

Clean Rebuild of T3000 - VS2019.sln (x86 / Release), 0 errors:

	Before	After
Total warnings	799	533 (−33%)
C4091 typedef ignored	253	4
C4715 not all paths return	25	12
C4717 infinite recursion	1	0
C4716 missing return value	1	0
C4172 dangling pointer	2	0
C4702 unreachable code	0	0
C4094 untagged struct	0	0

The last two rows are the regression check: no new warnings were introduced, and C4702 remaining at 0 confirms that none of the inserted return statements landed in a function that already returned on every path.

What changed

Two commits, deliberately separated so the risky and the risk-free parts can be reviewed differently.

1. Latent bug fixes (13 files) — can affect runtime behaviour:

• CWorkspaceBar::OnMouseMove called itself rather than the base class. Any mouse move over that pane was a guaranteed stack overflow.
• set_offline_mode is declared bool and dllexported but never returned, so callers read a junk register. All five callers ignore the result, so the fix preserves behaviour and keeps the exported signature.
• intervaltotextfull and intervaltotext_old both returned a pointer to a 12-byte stack local. Every caller passes an output buffer and ignores the return value.
• 13 missing return statements on paths that fell off the end of a non-void function.

2. typedef cleanup (6 files) — provably no-op: 17 declarations like typedef struct Foo { ... }; have no alias after the closing brace, so the compiler was already discarding the keyword.

Notes for the reviewer

Return values were chosen conservatively. Each follows a convention already present in the same function or API — mudbus_read_one returns -2 on retry exhaustion to match its sibling loop; S_OK for the WebView2 completion handler; 0 for the MFC ON_MESSAGE handler.

Ten C4715 sites were deliberately left unfixed. Where the correct return value could not be determined from the code, guessing would convert today's non-determinism into silent, deterministic incorrect behaviour — strictly worse. These need someone with protocol knowledge:

• Post_Refresh_Message is the one to look at. Its MODBUS branch has its entire body commented out including the return TRUE/FALSE. That reads like an unfinished edit and may be a real functional bug rather than just a warning.
• WritePrivateBacnetToModbusData (×2, T3000 and ISP copies) — is the timeout path 0 or -1?
• ShutDownMstpGlobal has a bare return; in an int function.
• Plus isbacnet_point, check_bacnet_data, Load_Product_Value_Cache, GetPriavteDataByPanelBlocking, ReadDataAgain, WritePrivateBacnetToModbusCharData.

Two further issues found but not touched here:

• dlmstp.c:744,749 pass narrow string literals to CreateSemaphore in a Unicode build (C4133 ×2), so both semaphore names are garbage. A _T() wrap away, but it belongs in its own change.
• A third copy of intervaltotext_old with the identical dangling-pointer bug lives in T3000ProgramMoudle/dllmain.cpp:1630. Left alone because that project is not referenced by this solution and therefore isn't built.

Only declarations closing with a bare }; were touched in commit 2. Ones that do define an alias (e.g. } URLStruct;) were left alone — stripping typedef there would turn a type alias into a variable definition.

Unrelated, but worth knowing: upstream CI has been failing since ~2026-06-10, when GitHub's windows-latest moved to Visual Studio 2026. The runner lacks C++ MFC for v143 and the .NET Framework 4.5.2 targeting pack, producing MSB8041/MSB3644. That predates this branch and is not affected by it.

🤖 Generated with Claude Code

[dieseld23]

Opened against the wrong repository by mistake -- this was intended for the dieseld23 fork, not upstream. Closing. Apologies for the noise.