Downgrade entry point in library error to warning

[rmshaffer]

Is your feature request related to a problem? Please describe.
After discussion with @bettinaheim, we believe that the issue microsoft/iqsharp#266 is due to the fact that IQ# is discarding certain compiler errors and attempting to build an assembly anyway, but the compiler is not executing the rewrite steps after the error has occurred.

Describe the solution you'd like
We would like a new option in CompilationLoader.Configuration by which we can tell the compiler to execute the rewrite steps anyway, even if there were errors in compilation.

Here is the code in IQ# where this new option would be consumed:
https://github.com/microsoft/iqsharp/blob/186bacb0570ef8688ffa8dac48d6a1230169b61f/src/Core/Compiler/CompilerService.cs#L129-L138

Describe alternatives you've considered
There are other options for IQ# to work around the particular compiler errors that are being discarded, but simply compiling and then discarding the errors is the approach that requires the least low-level code manipulation on the IQ# side, and which has been working fine for the last couple of months.

[ScottCarda-MS]

I don't think it is a good precedent to set to allow the compiler to create a compilation despite errors having been thrown. This leaves the compiler open to having to support use cases for the compiler in which erroneous compilations occurred, which is not a situation we want to be in. It also biases future development work on the compiler to not have as strict a standard when considering what should and shouldn't be errors in the future, if users are capable of ignoring the error anyway. I'm concerned that this is a bad "easy" solution to the problem at hand with the IQ#, and that we should use this opportunity to come up with a better solution.

[bamarsha]

The reason that rewrite steps don't execute when there are errors is because they assume that the syntax tree they are given is valid; for example C# generation was crashing with an exception when it encountered invalid identifiers, causing confusing error messages.

I think if we are considering this we need to make it clear that rewrite steps can receive invalid ASTs, and make sure all of our built-in rewrite steps can handle any syntax tree gracefully (maybe with a precondition to skip the rewrite step if the AST is invalid).

But mostly I agree with @ScottCarda-MS. In general, errors are not supposed to be optional. Wouldn't it be better to make this particular error about entry points in libraries into a warning?

[rmshaffer]

I agree, that also sounds like a reasonable solution and doesn’t set a bad precedent about ignoring errors. Feel free to repurpose this issue for that if you feel that’s the best way to address this.

[rmshaffer]

@SamarSha @ScottCarda-MS Do you think there would be any issues with making the change to downgrade "entry point in library" to a warning? Would it be reasonable for me to raise a PR for that?

[bettinaheim]

@rmshaffer, @SamarSha, @ScottCarda-MS Rewrite steps indeed should only deal with valid compilations. Making it a warning instead of an error I think is a good solution in this case. I'll update the title accordingly, and we can possibly take a look at that for October.