• Sean Kelly <sean invisibleduck.org> Oct 13 2009
• Michel Fortin <michel.fortin michelf.com> Oct 13 2009
• Sean Kelly <sean invisibleduck.org> Oct 13 2009

Sean Kelly <sean invisibleduck.org> writes:

MIURA Masahiro Wrote:

 Jeremie Pelletier wrote:
 I also don't believe one model is "ruling them all".


 Let me clarity this, just in case I have caused an unnecessary
 confusion:  I think Sean's Erlang-like API is meant to coexist
 with the current core.thread, and that you can use one or both
 of them to build higher-level concurrency models.


Exactly.  That's also why I settled on the Erlang-like API instead of something
like coroutines.  It's possible to build a coroutine API on top of the Erlang
API, but not vice-versa.

 I think it's nice to have core.thread *and* message-passing API
 in Phobos.  Thread alone is too primitive for daily use, but
 we don't want to have too many concurrency models.


I see core as holding language-essential stuff and "close to the metal" APIs,
and have no inclination of hiding core.thread or core.sync from the user.  If
someone wants to work at this level and they understand the risks they should
be able to.  Any higher-level stuff like the message-passing API I proposed
would live in Phobos and sit on top of what's in core.

I disagree about poor performance though.  With unique references or move
semantics, a copy of even complex data isn't necessary to ensure that a message
is passed safely.

Michel Fortin <michel.fortin michelf.com> writes:

On 2009-10-13 11:39:21 -0400, Sean Kelly <sean invisibleduck.org> said:

 I disagree about poor performance though.  With unique references or 
 move semantics, a copy of even complex data isn't necessary to ensure 
 that a message is passed safely.


Yeah, but for unique reference to be usable you need lent semantics, 
otherwise you can't make sure you're the unique holder once you call a 
function to do something with your unique object.

Anything that use the reference could be storing it elsewhere:

	unique!Object o = new Object;
	o.doSomething();

How in the example above can I enforce that doSomething() won't escape 
the 'o' reference elsewhere? 'doSomething' could be made 'pure', but 
that trick will only work for one-argument functions (in the case 
above, the argument is the implicit 'this') because otherwise a pure 
function could leak the reference through another argument.

-- 
Michel Fortin
michel.fortin michelf.com
http://michelf.com/

Sean Kelly <sean invisibleduck.org> writes:

Michel Fortin Wrote:

 On 2009-10-13 11:39:21 -0400, Sean Kelly <sean invisibleduck.org> said:
 
 I disagree about poor performance though.  With unique references or 
 move semantics, a copy of even complex data isn't necessary to ensure 
 that a message is passed safely.


 Yeah, but for unique reference to be usable you need lent semantics, 
 otherwise you can't make sure you're the unique holder once you call a 
 function to do something with your unique object.
 
 Anything that use the reference could be storing it elsewhere:
 
 	unique!Object o = new Object;
 	o.doSomething();
 
 How in the example above can I enforce that doSomething() won't escape 
 the 'o' reference elsewhere? 'doSomething' could be made 'pure', but 
 that trick will only work for one-argument functions (in the case 
 above, the argument is the implicit 'this') because otherwise a pure 
 function could leak the reference through another argument.


I honestly don't know how to enforce this-- I simply mentioned it because
people have suggested it.  Phobos already has assumeUnique() for converting to
invariant, which is one option.  I was hoping someone could suggest
alternatives.