feature netcore3 support

[jogibear9988]

todo:

• documentation does not compile
• nrefactoy in wpf build

[dgrunwald]

I have no idea why the weak reference tests started failing.
Either something in WPF changed and introduced a memory leak; or something in .NET changed regarding the GC, and we can no longer test that an object gets collected as expected.
Oh, and just now I remembered: it's also possible that the test itself is keeping stuff alive. I remember having problems like that with DocumentDoesNotHoldReferenceToLineMargin, e.g. because object initializers involve an extra temporary that maintains a reference. Maybe the problem is just that the textView = null; started getting optimized out? Possibly by Roslyn, or possibly by updates to the .NET JIT.
I guess a clean test would have to use the approach from DocumentDoesNotHoldReferenceToLineMargin and reference the object-to-be-collected only inside a separate method (which probably should be marked [MethodImpl(MethodImplOptions.NoInlining)] for extra safety).

[jogibear9988]

Inside of Visual Studio the test works. (on my system). Von meinem iPhone gesendet

…

Am 05.05.2019 um 12:30 schrieb Daniel Grunwald ***@***.***>: I have no idea why the weak reference tests started failing. Either something in WPF changed and introduced a memory leak; or something in .NET changed regarding the GC, and we can no longer test that an object gets collected as expected. Oh, and just now I remembered: it's also possible that the test itself is keeping stuff alive. I remember having problems like that with DocumentDoesNotHoldReferenceToLineMargin, e.g. because use object initializers involve an extra temporary that maintains a reference. Maybe the problem is just that the textView = null; started getting optimized out? Possibly by Roslyn, or possibly by updates to the .NET JIT. I guess a clean test would have to use the approach from DocumentDoesNotHoldReferenceToLineMargin and reference the object-to-be-collected only inside a separate method (which probably should be marked [MethodImpl(MethodImplOptions.NoInlining)] for extra safety). — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[dgrunwald]

If it's a compiler optimization, it might depend on debug/release.
If it's a JIT optimization, it also might depend on whether a debugger is attached.
So that would indicate that it might be a problem with textView = null; getting optimized out.

[jogibear9988]

i think with a new major it‘s okay to drop 3.5 support.
if we encounter a critical bug we could backport a fix if needed

[dgrunwald]

• I don't know enough about the .NET core stuff to usefully review the build system changes.
• WeakReferenceTests: I can try changing these tests to be less susceptible to the GC keeping stuff alive for longer than necessary, by moving stuff into separate methods.
• Removing [Obsolete] stuff: OK, we can do that in a new major version; though ideally only for stuff that was already marked obsolete in the 5.0 release.
• Removing .NET 3.5: OK
• Removing the WithNRefactory build: OK

[dgrunwald]

Note: the readme should be updated to explain what to install, so that AvalonEdit can be built.
In particular when opening the solution in VS requires unusual steps like installing a preview version of a .NET SDK...

[christophwille]

@dgrunwald I have amended readme.md with a short section on How to build.

WeakReferenceTests: would be great if you could try to fix those - having full unit test coverage back would be very reassuring.

[christophwille]

I have added a separate issue for removing old code (#if NREFACTORY see #174) - with that, anything missing for merging this? Tests run again thanks to Daniel, SHFB doc generation works again, all artefacts are produced (nupkg, snupkg, source code links)