15-Feb-2013 00:18, Jacob Carlborg пишет:
On 2013-02-14 21:14, Dmitry Olshansky wrote:
I believe that the idea is that any amount of helpers is fine as long as
they are private and don't obfuscate code. Phobos contains a lot tiny
helpers that _might_ be useful but not exposed because of dubious
general utility and potentially confusing names.
As far pushing good primitives into Phobos, it'd better be done one step
at time and decoupled of the actual inclusion of Orbit (if we can agree
on it being included) into say tools repo and being bundled with DMD.
I agree that might be the best idea. But that is mostly for the "any"
function and similar smaller utility functions. The bigger problems are
like the serialization modules. Sure they could be put directly in Orbit
as well. But I still got dependencies on Tango for stuff like XML, Zip,
net and argument parsing.
Then there are these ways forward IMHO:
a) Admit that tango for D2 exists (easy) and bundle it with DMD (the
hard/not likely/inconvenient part)
b) Agree that we need to port it and issue a call to port/re-write
required facilities for Orbit on top of phobos/curl. This means pulls
against Orbit repo not phobos BTW.
c) Forget about Orbit and try something else, like Dub?
About serialization (that seems the biggest roadblock) - what exactly
does Orbit need it for? Maybe it can be decoupled and/or easily
re-written by hand until Orange or similar stuff gets into Phobos.
Regardless I think reducing dependencies is the important for inclusion
of any new component into the "D core".
--
Dmitry Olshansky
