Set up CI testing for windows.

[rnburn]

AppVeyor was set up for the repository (See #80), but the build tests still need to be written.

[mdouaihy]

I am working on it. I should have something in few hours.

[cwe1ss]

Just let me know if you need anything from the AppVeyor account!

[hypnoce]

@mdouaihy you found the issue for ctests ?

[mdouaihy]

it's taking a bit more time than expected because MSVC handle the shared libraries differently than Linux. Also, MSVC requires to declare the exported and imported symbols using declspec(export) and declspec(import) while compiling the shared lib but not when compiling the static lib, etc. I should have a working version running and tested on Appveyor by the end of this week end

[rnburn]

@mdouaihy - how's this going?

It looks like your branch also adds dynamic loading support for windows -- that's cool; but could we maybe break that out into a separate PR and put in a smaller PR that just adds windows CI support without dynamic loading to get started?

[mdouaihy]

@rnburn, I took time before submitting any PR because in parallel, I worked on building Jaeger on Windows also (https://github.com/mdouaihy/jaeger-client-cpp).

As for your suggestion, here are the problems:

• Unit tests were not building
  ** The first change i made fixed two compile errors build does not fix the link of the unit tests
  ** With Visual Studio, dynamic libraries should be moved next to the unit test binaries
• Unit tests are failing.
  ** unordered_map does not have the same order between MSVC and GCC/Clang
  ** there is a unit test for dynamic loading. If I dont add the dynamic support, the test will fail;

[hypnoce]

Great work from @mdouaihy who worked hard on it !

[rnburn]

@mdouaihy - thanks for the update.

This check should disable the dynamic loading test if it's not supported. Is it not working?

Which test does the unordered_map cause problems for? I can try to fix it.

[mdouaihy]

It's the test TEST_CASE("json"): The problem is that SpanData contains a tags which is an std::unordered_map<std::string, Value and SpanContextData contains baggage which is std::unordered_map<std::string, std::string>. The order is not guaranteed.

As for the Dynamic loading, you're right about the cmake check. However, I needed it while building Jaeger. That's why I included it in the change.

Just to be sure, the code is in the branch windowsbuild the repo I forked https://github.com/mdouaihy/opentracing-cpp/tree/windowsbuild

[rnburn]

@mdouaihy - I put in #92 to fix the JSON ordering issue.