Code in master has long been superseded internally, using code in next brings us the following benefits:
Natural support for more languages, from design perspective (code itself, protocol, source file layout);
Support for --coverage (i.e., gcov);
More backends for cache.
...
Yet, IIRC, it does regress in performance when compiling C++. This is mostly due to switch of protocol between client and daemon, from a dirty-and-quick one to a JSON based one. JsonCpp costs us quite some cpu cycles here.
That regression shouldn't hurt much, it's ~2%, and should be unnoticeable most of time.
Code in
master
has long been superseded internally, using code innext
brings us the following benefits:--coverage
(i.e., gcov);Yet, IIRC, it does regress in performance when compiling C++. This is mostly due to switch of protocol between client and daemon, from a dirty-and-quick one to a JSON based one. JsonCpp costs us quite some cpu cycles here.
That regression shouldn't hurt much, it's ~2%, and should be unnoticeable most of time.