Modernize library, fix protocol and connection bugs (v2.0.0)

[archofthings]

Summary

This brings the repository up to date: it fixes several real bugs in the Ethernet and serial protocol code, moves off the unsupported netcoreapp3.1, adds CI, and rewrites the README. The package version is bumped to 2.0.0 because a few public helpers became internal (details below)

Bug fixes

• Ethernet read requests were malformed. ReadWordAsync and ReadBitAsync wrote the Length field as 1 byte instead of 2, which shifted the whole data block by one byte. Per the CIMON master frame layout (ID(9) · FrameNo(1) · Cmd(1) · Res(1) · Length(2) · Data · CheckSum(2)), read requests now use the same 2-byte length as writes.
• Auto-connect failed after the first request. TcpSocket closed its socket after each request and then tried to reuse it, so with the default autoConnect: true every later call threw ObjectDisposedException. A new socket is now created for each connection.
• Frame numbers didn't wrap correctly. After 128 requests the frame number stayed at 0 for the next 128 requests. It now cycles 0–127.
• SerialSocket.ReceiveData could return a null task, which throws NullReferenceException when awaited.
• Truncated or corrupted responses threw IndexOutOfRangeException. They're now validated (header, frame number, length and checksum) and return ResponseCode.WritingError.
• Serial WriteWordAsync accepted up to 63 words, but the 2-character length field can only describe 61. Larger requests silently sent a wrong length and are now rejected.
• Opening an already open SerialSocket threw an exception.

Improvements

• Faster requests: responses are returned as soon as a complete frame arrives. Before, every request waited the full read timeout (1 s by default).
• Connection handling:
  • A connection opened explicitly with Connect() now stays open. Only connections opened automatically are closed after the request.
  • Requests on one connector are sent one at a time, so a connector can be shared between tasks.
  • TcpSocket has an optional usePing parameter for networks that block ICMP.
  • Connectors and sockets implement IDisposable.
• Refactoring: frame building and response handling are shared instead of copied into each of the 8 read/write methods.
• Platforms: targets netstandard2.0 and net8.0 instead of the unsupported netcoreapp3.1, so the package also works on .NET Framework 4.6.1+.
• Dependencies: removed Ardalis.GuardClauses. Its newer versions conflict with the custom extensions this project put in its namespace.
• Packaging: dotnet pack used to fail on any machine but the author's because the package icon pointed to C:\Users\...\cimon_logo.png. Packing now works and includes the README and symbols.

Breaking changes

• Tools and the guard extension methods are now internal.
• The PlcConnector constructor is now protected PlcConnector(IPlcSocket socket, bool autoConnect), and the protected _timeout field was removed.
• Serial WriteWordAsync accepts at most 61 words (was 63).

Tests and CI

• Tests now run on .NET 10 with current versions of xUnit, the test SDK and coverlet.
• The tests used async void, which in xUnit 2 can hide failures. They now use async Task.
• Added 14 regression tests for the fixes above: frame numbering, the read frame layout, reconnecting after auto-disconnect, keeping explicit connections open, and truncated, corrupted and missing responses.
• CI:
  • Replaced the two duplicate CodeQL workflows (still on v2 actions) with one current workflow.
  • Added a CI workflow that builds, tests and packs the library.

Docs

• Rewrote the README:
  • NuGet, downloads, CI, CodeQL, framework and license badges
  • Quick start, API overview and connection handling
  • Memory type, response code and request limit tables
• Added a LICENSE file (MIT, matching the package metadata) and a CHANGELOG.md.
• Package URLs now point to archofthings/Cimon.Net.

Testing

• dotnet build: no warnings for netstandard2.0 or net8.0
• dotnet test: 149 passed, 0 failed
• dotnet pack: the package is created, with the README and symbols
• Not tested on a real CIMON PLC

Not changed, worth checking

Serial WriteBitAsync sends each bit as a raw 0x00/0x01 byte rather than the ASCII characters '0'/'1'. That's unusual for an ASCII protocol, but I couldn't confirm what the spec expects, so the behavior is unchanged.