Survived Memory

[timcassell]

In an attempt to address my question on #1579, I worked on this.

I added an optional includeSurvived flag on MemoryDiagnoserAttribute and --memorySurvived command line option to capture the survived memory from a benchmark run.

[dnfadmin]

All CLA requirements met.

[timcassell]

.NET Core 3.1 and 5.0 measurement issue: dotnet/runtime#45446

[timcassell]

Looks like measurement errors are fixed in .Net 7 as of dotnet/runtime#67160.

[JaggerJo]

Looking forward to this feature 👀

[timcassell]

Unfortunately, survived memory is not so simple for the framework to measure automatically. The first invocation will include jit allocations and static memory, and the second invocation will miss pooled objects. When I used this in my own benchmarks (https://github.com/timcassell/CSharpAsynchronousBenchmarks), I took care to measure only pooled objects by adjusting the benchmarks in a complicated way. I don't think it's good for users to have to do that to get useful measurements. I think this will be better served by #784, so users can capture the survived memory that matters to them.