Ternary operators, code optimization, and branch coverage

[leighshepperson]

Ternary operators, code optimization, and branch coverage

[SteveGilham]

Should be fixed in release 4.0.649.

[SteveGilham]

If altcover has an underlying assumption here, it's going to be that the IL it's processing looks more like what the F# compiler emits than a Roslyn compiler's output, rather than being a preference for debug over release.

The Gendarme module is used to compute method cyclometric complexity -- it's a re-implementation of the part of the Mono Gendarme library that OpenCover uses for the same purpose that exists only because there isn't a .net core version of the original available. That code may well have a tacit assumption that it would only be used on debug builds.

The code injection for branch coverage detection is intended to be agnostic as to how the IL was composed, while not triggering false positives for the many common cases where the compiler injects its own branches that only transfer control within a sequence point.

[SteveGilham]

The code of interest is of the form return selection ? : yes : no; which compiles to IL which returns directly from that sequence point, wherever it is found in a method. It seems to be a fairly recognisable pattern

 brtrue.s label
...
ret
label: ...
...
ret

so that should be fairly simple to add a special case for.

[leighshepperson]

Yes, great - that addition will be extremely helpful.

[SteveGilham]

Pre-release build 646 contains the substantive change for the branch coverage, though not for all the related measures e.g. complexity.

[leighshepperson]

Great - thank you for providing a fix, it'll be a great help