Enabling macOS compilation & compatibility

[linkineo]

• inglued was not compiling under macOS (v10.12.3), as the "process" library was not compatible (missing feature.h). This has been corrected here so I integrated it. Also, it seems you did not add this dep as a subtree (or maybe I did not get the history :-).
• inglued was not working under macOS as a known bug in nlohmnan-json for macOS prevents proper parsing of the inglued file. I integrated V2.0.5 as a subtree. Previously integrated version seemed older (no version number), and had a runtime error when parsing inglued file ('std::invalid_argument: parse error - unexpected '). Did not integrate later versions as they have other unresolved issues with macOS (std::invalid_argument: stream error). Also, using subtree for the json library is heavy as the repo is quite rich. Maybe we could revert to just using the header file.

[daminetreg]

Hi Patrick,

Thanks you for your change. :wink: I welcome them, I wanted to check for the macOS support yesterday but didn't got the time.

Also, it seems you did not add this dep as a subtree (or maybe I did not get the history :-).

Yes this project has to be inglued by himself, but wasn't yet the case. As when I started building inglued I didn't had inglued to deal with the deps. I aim to run inglued on himself in the future.

This was a first development in a pragmatic way to simple strike through. But I find it good you started cleaning it up.

However, I have 1 problem with that : boost filesystem is not header only, so I was weighing 2 options :

• move to c++17 the codebase so that I can benefit from std::filesystem
• make boost filesystem header only by some hardcore automated machinery

inglued was not working under macOS as a known bug in nlohmnan-json for macOS prevents proper parsing of the inglued file. I integrated V2.0.5 as a subtree. Previously integrated version seemed older (no version number), and had a runtime error when parsing inglued file ('std::invalid_argument: parse error - unexpected ').

2.1.1 was last release, why not this one ? I don't know which version I added, I think I took the one I just had checked out, for the same reason explained above.

Also, using subtree for the json library is heavy as the repo is quite rich. Maybe we could revert to just using the header file.

This is something actually inglued could solve, but the implications have to be checked in detail, if we decided to implement a commit-ammending git-subtree operation which would only keep the header file pointed to by the "include_path" info or "project_srcs" infos we would reduce seriously the heavy-weightness of repositories that we inglued. Like nlohmann json which has an 128mb test input file.

Perhaps we should do that ? But then one cannot use git-subtree command himself, which is not necessarily a requirement of inglued. What is your opinion about that ?

I'm giving a try to your changes on my macOS VM, if they work for me I'll merge.

[daminetreg]

@linkineo I opened #2 as well, as the other discussion above, if you like to work on #2 tell me, so that we don't collide today evening.