On 05/16/21 18:05 PM, Stuart Henderson wrote: > On 2021/05/16 09:51, Anthony J. Bentley wrote: >> >> This packages 1.1.0 which was released in 2016. But rapidjson sees >> continuous development in Git, just no releases; see >> https://github.com/Tencent/rapidjson/issues/1759 >> >> As a result, many projects embed newer versions. Are we sure having >> this installed won't break any existing ports?
Sorry, I didn't check if any other ports were already including this as a dependency. I found the following using grep in /usr/ports, but I guess that will miss some: devel/leatherman : bundles 1.0.2 in the upstream tar emulators/mgba/ : bundles an unknown version in the upstream tar games/barony : fetches release 1.1.0 in the port makefile games/ja2-stracciatella : bundles 0.11 in the upstream tar net/bro : bundles an unknown version in the upstream tar > > I agree with your concerns. Also ccls uses a specific commit id of > rapidjson as a git submodule: > https://github.com/MaskRay/ccls/tree/master/third_party > Yes, although CMakeLists.txt checks for minimum version 1.1.0 if USE_SYSTEM_RAPIDJSON is true. That might be because Linux distros which aggressively de-vendor dependencies are packaging that. I'm happy to bundle it in ccls and drop the separate port if that's preferred? Or ccls can fetch it directly like games/barony does. > Nick, I think you should just to generate a proper tar for ccls rather than > using the broken github autogenerated one. Sure. But does it need to be hosted permanently somewhere? > > (Also wondering, is ccls going to be another port that is going to need > work every time clang is updated?) I don't know, but I made this on 6.9 originally with clang/LLVM 10 and then rebuilt it without modification after updating to current with 11. -- Thanks, Nick
