Drive the solution to zero warnings / zero errors with analyzers on,
promote the consistently-fixed rules to error, and lock in coverage
on the only line of new logic.
Generate stack and application code using source generators
Change built in types to readonly structs
Update localized text implementation to use JSON text serializer
Status code with symbolic names
Immutable node ids
No boxing for all built in types <= 8 bytes in Variant.
Split user identity token and token handling via token handler
Add migration guide for breaking changes
This pull request modernizes the OPC UA .NET codebase and tests, removes .NET Standard 2.0 support from certificate/security code, and updates test code to use C# 12 and .NET 8+ features for clarity and maintainability.
- Certificate and BouncyCastle interop files (e.g., CertificateBuilder.cs, X509Extensions.cs, X509SignatureFactory.cs) are refactored to drop NETSTANDARD2_0 support and use modern .NET APIs and conditional compilation.
- Test code across the codebase is updated to use default, strongly-typed Parse methods, collection expressions, and improved async/await patterns.
- Exception handling and assertions in tests are modernized, and test method naming is made consistent with Async suffixes.
- The project and build files are updated to clarify Roslynator analyzer usage and remove obsolete netstandard tags.
- General code and test cleanup includes improved null safety, type safety, and removal of deprecated or unused code.
This change adds a **new Opc.Ua.Types** nuget that contains core types that do not rely on generated code. Why the name? Because almost all code that was in the Types subfolder is part of it, and because we do not want to change the name of the Opc.Ua.Core nuget/assembly because code relies on it. Any other name like base or bootstrap felt out of place due to the current structure already defining the name "types". Key design is to depend on only a small set of libraries with limited transitive dependencies, that includes hashcode, logging.abstraction, and for non standard20 builds System.Text.Json (needed for localizedText mul parsing). Also, key is to support .netstd 2.0 to support source generators and to allow removal of EEC_SUPPORT defines by removing std2.0 from core and all libraries.
The changes are mostly restructuring, and thus small but contain some key changes:
- **Status code** struct now comes with a Symbol field, which is used to carry symbols in case the code is not generated. Currently (until generated) the status code struct must be manually created with a symbol or the ToSymbolicId extension must be called in the core project. Future will generate Status code constants just like NodeId that contain the symbol which also removes the reflection-based lookup. Several tests needed to be adjusted as they checked the status code symbol strings instead of the code.
- **Core identifiers** are copied as they were needed for code to compile, that includes attributes, a subset of status codes and node identifiers. These will never change so no need to generate them.
- The types folder now contains **encoders**, except for json which is still in core. Once the json encoder is ported to system.text.json it will also be included and Newtonsoft.Json dropped as dependency (-> #3329).
- Since types relies on **IOperationContext** and **ISystemContext**, and these export IUserIdentity which relies on crypto, these were split and extended in core with ISessionOperationContext and ISessionSystemContext, which mean changes to ServerSystemContext and introduction of SessionOperationContext and SessionSystemContext classes that extend OperationContext and SystemContext. Some runtime checks in the server are therefore needed to get at the session and auth info, however, those are anyway optional fields and are not on the critical paths, and thus a type check is in order (and was done in the past already, but against OperationContext class).
- **Manifest resources** are now split between Core (nodeset zip, some generated resoures) and Types (built in bsd/xsds). This needs to be revisited later if the bsd/xsd need to be regenerated during type generation. Note that the embedded resource names changed as a result. Since these are only used internally and do not constitute public API this feels reasonable vs duplication of resources in 2 assemblies.
- Reflection however now needs to load not just the built in **encodeables** from Opc.Ua.Types assembly, but also the generated encodeables from Opc.Ua.Core assembly. For this a new helper Utils method was added to get the Core assembly when linked against Opc.Ua.Core otherwise null.
- Whatever utils methods needed to be split out into the new Opc.Ua.Types assembly, they are now in **CoreUtils** static class in Opc.Ua.Types, and Utils.xxxx forwards to them to ensure minimal changes. Later we would want to obsolete them in favor of CoreUtils, although it might make sense to split them again into domain specific utilities, such as networking, files, etc
Marc Schier