On 2021/04/09 07:19, niamkik wrote: > Hi, > > > > Since I pruned the old erlang versions in ports, I've got to ask. Would > > we get sufficient benefits from keeping more than one erlang version? Do > > we have important erlang software people rely on that we can't build > > with a single chosen version? > > > > We routinely make decisions choosing single library versions. It makes > > sense to lean this way for the compilers...
Libraries are often difficult to handle with multiple versions, they often conflict and cause problems if several are installed together. Though in some cases it's fine, e.g. we often have multiple versions of gtk+, gegl, gstreamer in parallel, but it does require some thought. For interpreters there's often good reason to have multiple versions (see php, python, lua, ruby) and *if* there's a good reason and somebody is willing to maintain then I wouldn't object to that. (That latter point is key though, if it's not going to be actively maintained then it really wants keeping simple). > On my side, maintaining 4 versions will not be easy but I can understand to > have backward compatibility with older version, it could be nice. Anyway, > maintaining only one release could be easier, like R22 or R23. It will > probably lead to break dependencies though. Most of the dependencies are either already broken (riak, rabbitmq), or can handle at least up to 23 (rebar3, elixir). The others are rebar (no longer maintained and only used in ports as a dependency of riak) and tsung (no indication whether it will work or not and doesn't exactly seem active upstream). (riak is a bit special anyway, seems they really want it to be built with a patched Erlang..)
