On Saturday, 2 January 2016 at 04:27:02 UTC, Jack Stouffer wrote:
On Friday, 1 January 2016 at 20:10:22 UTC, Dicebot wrote:
Idea of any "automatic" inclusion is very harmful because by
current standard std.experimental modules are not required to
have Phobos quality and can be very far from being good enough.
My proposal was based on the quality of std.logger and
std.allocator when they were merged. I wasn't aware that
modules that were unfinished could be merged into
std.experimental.
IIRC, the review for std.experimental.logger specifically left
some stuff hanging with the idea that it would be sorted out in
experimental, but I could be remembering wrong. Regardless, a
certain level of quality is required for something to get into
std.experimental, because it has to pass the review and vote, but
that review and vote is no longer done with the expectation that
what's there is necessarily set in stone when it gets merged
(otherwise, there's no point to std.experimental). It should be
solid, but there is now the potential for redesign on some level
before it gets into std, whereas before, it would have just gone
into std, and we basically had to get it right up front.
Having another review for something like std.experimental.logger
before it gets merged into Phobos makes sense, because it would
allow us to have an open discussion on anything that needs to be
changed now that folks have been using it. And if it's ready to
go (potentially decided by a vote), then we could merge it. But
without that review, the only stuff that will have changed is
stuff that gets reported in bugzilla or that a Phobos dev
specifically decided to change. If we want to actually find out
how well it's been working for folks, then we need a way to get
feedback based on actual usage and discuss it. Another review
doesn't necessarily have to be the way that we do that, but it's
probably the best mechanism that we have.
- Jonathan M Davis
