AltCover - 5 years on : what happened in 2022

2022 was a comparatively quite year, only 8 releases in all, mostly bugfixing, tracking compiler changes, or working around places where I put Mono.Cecil under unexpected stress, and have to work around until it gets a next release; but with a lot of changes in processes, under the covers. You can read the release notes for the details; this post is aimed at the bits that the end user shouldn't see.

The year started by getting my FxCop wrapper into a state where I could persuade it to process netstandard2.0 assemblies, at least most of the time, thereby enabling me to drop the debug-only net472 builds for libraries. Not all net472 builds could in fact be processed - the Avalonia-based GUI threw intractably even though it could build happily against that target -- The following error was encountered while reading module 'System': Security attribute type does not have a default constructor.

Allowing assembly-level suppression attributes meant that I could also enhance the Gendarme-based scrutiny of the codebase; even if sticking at netstandard2.0 means warnings about not using preferred method overrides from later API levels.

The release of Cake v2 gave me excuse to test and update the relevant APIs -- and another such overhaul in the autumn with Cake 3.0 dropping the deprecated NetCore names.

This was the year to also add dependabot, mostly for keeping the CI build actions up to date, with some prompting about other dependencies; though things like F# updates with compiler changes were actually the most significant unexpected maintenance-level change.

As a maintenance tool, fantomas matured this year, especially with its integration of .editorconfig, to the point that there are only a few "quirky" files where I have to back changes out, mostly with hacky #if/#endif nesting; so it's an on-going tool rather than a once-per release spring-clean. The worst bit is when a compiler generated method with a line-number in its name gets moved, and the static analysis suppression shutting up warning about the bad habits of code generators has to be updated.

The already massive Fake build script also increased its level of automation, keeping track of SDK versions via XML parsing of project files, to keep static analysis dependency paths up to date.

Belatedly, after the release in late 2021, the icons for VS2022 became available, and have been incorporated into the visualizer; this was also the year when I dropped the abortive FuncUI stub, and instead started to experiment with the massive changes in Avalonia 11 - that will be something on-going into the coming year!

Having started to centralize build process late in '21 by using Directory.Build.* files, late '22 saw the adoption of central package versioning, which simplifies much of the update process across the multi-solution sprawl. On the upside, while there's only one place to update, the Visual Studio UI doesn't do that, so while it can identify the updates, it's still a manual process at that one point. It also means having to add more verbiage to Nuget.config files intended to consume just-minted packages as part of build validation testing.

The big change at the end was not net7.0 itself (problems with the 7.0.100 release on dotnet test aside), but the way that didn't play well with the Fake CLI tool, or with the Appveyor CI; so I've replaced the former with net7.0 built drivers for what were scripts and are now source files within the drivers; and the latter with GitHub for doing things like Coveralls reporting -- redone as the old tool I'd been using stopped working at some point mid-year -- and release to NuGet. Of course with my build versioning having incorporated the AppVeyor build number, I needed to take a gratutious minor version bump as even my GitHub build numbers were lower than the AppVeyor ones. In the end, I took a big break and have (ab-)used NerdBank.GitVersioning, starting with the 1-Jan-23 anniversary release.

In many ways the 7.0.100 fail was a forunate one, as it gave me a reason to improve the AltCover scripting APIs, used extensively by AltCover's own build process, to emit raw property/value pairs rather than just composed command-line arguments; and having dropped support for early Fake5 releases, to use its property/value APIs rather than the composed command-line ones.

One last little bit of build-process work was in sorting out the assertion behaviour of Unquote - it casts about to see what libraries are visible and makes an opinionated choice (with NUnit being at the end of its preferences); a little hacking to catch and re-assert means that I can now do Assert.Multiple of Unquote tests - or take multiple bites at a failure, when Unquote tells me that two long strings differ, NUnit can point at where.