[Bug libstdc++/42819] New: [C++0x] std::async fails to compile with simple tests, including N3000 example

[Bug libstdc++/42819] New: [C++0x] std::async fails to compile with simple tests, including N3000 example

[lloyd at randombit dot net]

Here is the example for std::async included in the N3000 draft (section
30.6.9):

#include <future>

int work1(int value);
int work2(int value);

int work(int value)
   {
   auto handle = std::async([=] { return work2(value); });
   int tmp = work1(value);
   return tmp + handle.get();
   }

With svn r156097, it fails to compile:

$ g++-4.5-r156097 -std=c++0x -c n3000.cpp 
n3000.cpp: In function 'int work(int)':
n3000.cpp:8:57: error: no matching function for call to
'async(work(int)::<lambda()>)'
n3000.cpp:8:57: error: unable to deduce 'auto' from '<expression error>'

It also fails with `auto handle = std::async(work2, value);`

$ g++-4.5-r156097 -v
Using built-in specs.
COLLECT_GCC=g++-4.5-r156097
COLLECT_LTO_WRAPPER=/usr/local/gcc-4.5-r156097/libexec/gcc/x86_64-unknown-linux-gnu/4.5.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: ../gcc-svn/configure --enable-languages=c,c++
--prefix=/usr/local/gcc-4.5-r156097 --program-suffix=-4.5-r156097
Thread model: posix
gcc version 4.5.0 20100120 (experimental) (GCC)


-- 
           Summary: [C++0x] std::async fails to compile with simple tests,
                    including N3000 example
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libstdc++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: lloyd at randombit dot net
 GCC build triplet: x86_64-unknown-linux-gnu
  GCC host triplet: x86_64-unknown-linux-gnu
GCC target triplet: x86_64-unknown-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #1 from paolo dot carlini at oracle dot com  2010-01-21 10:28 -------
Jon, what's going on, here? I'm referring first of all to the version without
lambdas, so:

#include <future>

int work2(int value);

void work(int value)
   {
     auto handle = std::async(work2, value);
   }

As far as I can see, the problem is with the typename _Fn::result_type used in
the return type, which obviously work2 lacks. Looks like there is a defect in
the N3000 draft.


-- 

paolo dot carlini at oracle dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jwakely dot gcc at gmail dot
                   |                            |com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[jwakely dot gcc at gmail dot com]

------- Comment #2 from jwakely dot gcc at gmail dot com  2010-01-21 10:35 -------
Yes, it's a defect (a concepts hangover) - the example is invalid according to
the async spec. I pointed it out to Lawrence and he's dealing with it for the
next draft.

I implemented async as defined in n3000 - I can change it to use result_of /
decltype if desired.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

-- 

paolo dot carlini at oracle dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |SUSPENDED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #3 from paolo dot carlini at oracle dot com  2010-01-21 10:37 -------
Agreed. Thus, let's suspend this for now.


-- 

paolo dot carlini at oracle dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2010-01-21 10:37:15
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #4 from paolo dot carlini at oracle dot com  2010-01-21 10:38 -------
Well, using result_of / decltype would make it more usable in the meanwhile,
should not be a big deal, right?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[jwakely dot gcc at gmail dot com]

------- Comment #5 from jwakely dot gcc at gmail dot com  2010-01-21 10:42 -------
(In reply to comment #2)
> Yes, it's a defect (a concepts hangover) - the example is invalid according to
> the async spec. I pointed it out to Lawrence and he's dealing with it for the
> next draft.

To expand on this:  std::async is defined in terms of Fn::result_type, assuming
that Concepts would provide Callable::result_type, so the example would be
valid.
Without concepts, that doesn't work, so the example is invalid.
The most likely fix will be to use result_of<Fn(Args...)>::type instead, making
the example valid again.

I chose to follow the spec, not the example which appears in a non-normative
note.


(In reply to comment #4)
> Well, using result_of / decltype would make it more usable in the meanwhile,
> should not be a big deal, right?

Yes, I think we could do that.  I'll ask Lawrence how he's planning to address
it - we might as well do that way.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #6 from paolo dot carlini at oracle dot com  2010-01-21 10:45 -------
Thanks Jon.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[jwakely dot gcc at gmail dot com]

------- Comment #7 from jwakely dot gcc at gmail dot com  2010-01-21 10:54 -------
btw, Jack, thanks for testing it.  Your feedback is appreciated


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[lloyd at randombit dot net]

------- Comment #8 from lloyd at randombit dot net  2010-01-21 15:38 -------
Jon (and Paolo) - thanks for doing the work!

Is there an easy workaround I can apply locally for this? I tried replacing all
instances of `typename _Fn::result_type` with `typename
result_of<_Fn(_Args...)>::type` in future just to see if I can get things going
for me locally, but then get this error on the N3000 example:

In file included from
/usr/local/gcc-4.5-r156097/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../include/c++/4.5.0/future.2:38:0,
                 from n3000.cpp:1:
/usr/local/gcc-4.5-r156097/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../include/c++/4.5.0/functional:
In instantiation of 'std::result_of<std::launch(work(int)::<lambda()>)>':
/usr/local/gcc-4.5-r156097/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../include/c++/4.5.0/future.2:1328:41:
  instantiated from 'std::future<typename std::result_of<_Functor(_ArgTypes
...)>::type> std::async(_Fn&&, _Args&& ...) [with _Fn = work(int)::<lambda()>,
_Args = {}, typename std::result_of<_Functor(_ArgTypes ...)>::type = int]'
n3000.cpp:17:57:   instantiated from here
/usr/local/gcc-4.5-r156097/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../include/c++/4.5.0/functional:183:9:
error: 'std::declval [with _Tp = std::launch, typename
std::add_rvalue_reference<_Tp>::type = std::launch&&]()' cannot be used as a
function

I'm not sure if I have messed this up (my C++0x-fu is still very weak), or if
functional's result_of needs further changes to support this case - it looks to
be using decltype now (a recent change I think; thanks again!) so I had guessed
this would 'just work', but I had never even heard of declval and looking at
type_traits can't quite grok what it's doing or why this isn't happy. Simple
test cases I've tried using result_of on global functions and lambdas seem to
work so I'm having a hard time narrowing it down. Thus this comment. :)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #9 from paolo dot carlini at oracle dot com  2010-01-21 15:48 -------
Jack, I would ask you to have a bit of patience, because this code is in flux.
Just a few days. In the meantime, if you want to experiment, I would suggest
two things: 1- Always use up to date sources, r156097 is already old; 2-
Remember to update *all* the occurrences of that _Fn::result_type, also inside
the body of async(launch, _Fn&&, _Args&&...) for example; 3- Before spending
too much time in the interaction between lambdas and <future>, double check
lambdas per se, because we have some outstanding issues, in particular for
lambdas with no captures as function pointers.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[jwakely dot gcc at gmail dot com]

------- Comment #10 from jwakely dot gcc at gmail dot com  2010-01-21 16:08 -------
(In reply to comment #8)
> 
> Is there an easy workaround I can apply locally for this? I tried replacing all
> instances of `typename _Fn::result_type` with `typename
> result_of<_Fn(_Args...)>::type` in future just to see if I can get things going

That should be all that's needed, in theory.

> /usr/local/gcc-4.5-r156097/lib/gcc/x86_64-unknown-linux-gnu/4.5.0/../../../../include/c++/4.5.0/functional:183:9:
> error: 'std::declval [with _Tp = std::launch, typename
> std::add_rvalue_reference<_Tp>::type = std::launch&&]()' cannot be used as a
> function

I'm not sure what's going on here and can't look into it now. Does the example
in Comment 1 work?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #11 from paolo dot carlini at oracle dot com  2010-01-21 16:13 -------
No Jonathan, doesn't: that strange error happens (after having made public
explicit future(const __state_type&)).

I'd like only to understand if we have to ask Jason's halp again, or not...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[jwakely dot gcc at gmail dot com]

------- Comment #12 from jwakely dot gcc at gmail dot com  2010-01-21 16:14 -------
Aha - I see what's happening.

As well as changing _Fn:result_type everywhere please change line 1327 in
<future> to this:

    return async<_Fn, _Args...>(launch::any, std::forward<_Fn>(__fn),

The problem is the wrong overload of async() is being selected, and it is
trying to use launch::any as the functor, which gives the "cannot be used as a
function" error.

Paolo, this is the problem I had with std::bind, I think there might be a
larger problem with overloaded variadic template functions ... I'll think about
it further.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[jwakely dot gcc at gmail dot com]

------- Comment #13 from jwakely dot gcc at gmail dot com  2010-01-21 16:19 -------
(In reply to comment #12)
> Paolo, this is the problem I had with std::bind, I think there might be a
> larger problem with overloaded variadic template functions ... I'll think about
> it further.

To be clear: a problem in the standard, not g++ 

I'm starting to think that having std::bind and std::async as overloaded
variadic functions is a mistake. It would be easier if one overload was renamed
std::bind_r and std::async_any, as you wouldn't need to specify explicit
template arguments to disambiguate the overloads.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #14 from paolo dot carlini at oracle dot com  2010-01-21 16:21 -------
Great. Indeed, that way works. Actually, I was surprised to see result_of used
with launch named in the template argument?!?

Anyway, we also have the residual nit that I had to change to public the
explicit constructor future(const __state_type&). I suppose isn't a big
issue...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[jwakely dot gcc at gmail dot com]

------- Comment #15 from jwakely dot gcc at gmail dot com  2010-01-21 16:28 -------
(In reply to comment #14)
> Anyway, we also have the residual nit that I had to change to public the
> explicit constructor future(const __state_type&). I suppose isn't a big
> issue...

Did you change _Fn::result_type on lines 122 and 561?
async _should_ be a friend, and not need the public constructor


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #16 from paolo dot carlini at oracle dot com  2010-01-21 16:35 -------
I forgot the one at line 561, that explains it. Fine ;)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #17 from paolo dot carlini at oracle dot com  2010-01-21 16:57 -------
Given our current understanding of the issue, suspended status seems too much:
we can deal with it, reasonably soon, hopefully after Lawrence Crowl feedback.


-- 

paolo dot carlini at oracle dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|SUSPENDED                   |NEW
   Last reconfirmed|2010-01-21 10:37:15         |2010-01-21 16:57:08
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

-- 

paolo dot carlini at oracle dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |paolo dot carlini at oracle
                   |dot org                     |dot com
             Status|NEW                         |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #18 from paolo dot carlini at oracle dot com  2010-01-21 18:10 -------
Jon, I have to go now, I was testing the obvious patch but apparently we have
more of the same issue with std::bind, besides that in async(_Fn&&,
_Args&&...): the existing 30_threads/async tests are failing at compile time:
if you can give me an hint I can asjust the patch I have already and commit it
in, say, 5 hours from now. I think the problem now comes from the user
(testcase) code itself: there is an "ambiguity" between the two async: shall we
maybe use an enable_if on the first argument of async and check if the type is
launch?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #19 from paolo dot carlini at oracle dot com  2010-01-21 18:12 -------
... or via the usual trick of the defaulted template parameter.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[jwakely dot gcc at gmail dot com]

------- Comment #20 from jwakely dot gcc at gmail dot com  2010-01-21 18:20 -------
Eurgh - yes, we should disable one overload.  I'm getting less and less happy
about overloaded variadics.

I'll look into it later.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #21 from paolo dot carlini at oracle dot com  2010-01-22 02:07 -------
Created an attachment (id=19685)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=19685&action=view)
Draft patch

Jon, this version regtests fine, it's the simplest solution I have been able to
figure out so far. Please have a look, and in case commit it or any variant you
like, probably I will not be able to access svn until monday...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #22 from paolo dot carlini at oracle dot com  2010-02-05 12:55 -------
Jon, do we have some sort of reference / pointer for this issue? Are we sure
Lawrence is aware of it?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[jwakely dot gcc at gmail dot com]

------- Comment #23 from jwakely dot gcc at gmail dot com  2010-02-05 18:40 -------
I realised the other day I hadn't followed up on a fix for this.

Here's some of the email I got from Lawrence just before Christmas, I haven't
heard anything more recent:


> I noticed while implementing it that in n3000 std::async<F,...>
> returns std::future<typename F::result_type>
>
> Is the requirement on a nested typedef intentional, or is it a relic
> of an earlier version using concepts?

N3000 has a bad merge of two papers voted into the standard, so there
is much cleanup to be done.  Detlef and I will need address this issue
as well.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #24 from paolo dot carlini at oracle dot com  2010-02-05 18:46 -------
I see, great, just wanted to make sure the issue is recorded somewhere besides
this v3 PR.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [C++0x] std::async fails to compile with simple tests, including N3000 example

[jwakely dot gcc at gmail dot com]

------- Comment #25 from jwakely dot gcc at gmail dot com  2010-02-09 18:28 -------
This is now LWG 1315


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [DR 1315][C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #26 from paolo dot carlini at oracle dot com  2010-02-09 21:20 -------
Fine, let's suspend this, then.


-- 

paolo dot carlini at oracle dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |SUSPENDED
            Summary|[C++0x] std::async fails to |[DR 1315][C++0x] std::async
                   |compile with simple tests,  |fails to compile with simple
                   |including N3000 example     |tests, including N3000
                   |                            |example


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [DR 1315][C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #27 from paolo dot carlini at oracle dot com  2010-02-09 21:43 -------
Jon, about the proposed resolution, do you think that simple is enough? I mean,
it doesn't say anything about the problem I had to address with SFINAE in my
tentative patch...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [DR 1315][C++0x] std::async fails to compile with simple tests, including N3000 example

[redi at gcc dot gnu dot org]

------- Comment #28 from redi at gcc dot gnu dot org  2010-02-09 21:51 -------
Good point, I'll ask Howard to update it


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [DR 1315][C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #29 from paolo dot carlini at oracle dot com  2010-02-11 01:59 -------
Thanks to the recent fixes and all the good work Jason did, I'm pretty sure
that now a normal enable_if or decltype on the return type would work just
fine. I'm wondering if we should just do that, for 4.5.0 and allow people to
experiment more easily with these facilities... I don't think we are risking
much, in any sense, wrt the actual resolution of 1315...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [DR 1315][C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo at gcc dot gnu dot org]

------- Comment #30 from paolo at gcc dot gnu dot org  2010-02-12 22:31 -------
Subject: Bug 42819

Author: paolo
Date: Fri Feb 12 22:31:15 2010
New Revision: 156742

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=156742
Log:
2010-02-12  Jonathan Wakely  <jwakely.gcc@gmail.com>
            Paolo Carlini  <paolo.carlini@oracle.com>

        PR libstdc++/42819, DR 1315
        * include/std/future (async): Use std::result_of for the template
        argument of the std::future return type; adjust everywhere.
        * testsuite/30_threads/async/42819.cc: New.
        * testsuite/30_threads/packaged_task/cons/assign_neg.cc: Adjust
        dg-error line number.
        * testsuite/30_threads/packaged_task/cons/copy_neg.cc: Likewise.
        * testsuite/30_threads/future/cons/assign_neg.cc: Likewise.
        * testsuite/30_threads/future/cons/copy_neg.cc: Likewise.
        * testsuite/30_threads/promise/cons/assign_neg.cc: Likewise.
        * testsuite/30_threads/promise/cons/assign_neg.cc: Likewise.

Added:
    trunk/libstdc++-v3/testsuite/30_threads/async/42819.cc
Modified:
    trunk/libstdc++-v3/ChangeLog
    trunk/libstdc++-v3/include/std/future
    trunk/libstdc++-v3/testsuite/30_threads/future/cons/assign_neg.cc
    trunk/libstdc++-v3/testsuite/30_threads/future/cons/copy_neg.cc
    trunk/libstdc++-v3/testsuite/30_threads/packaged_task/cons/assign_neg.cc
    trunk/libstdc++-v3/testsuite/30_threads/packaged_task/cons/copy_neg.cc
    trunk/libstdc++-v3/testsuite/30_threads/promise/cons/assign_neg.cc
    trunk/libstdc++-v3/testsuite/30_threads/promise/cons/copy_neg.cc


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819

[Bug libstdc++/42819] [DR 1315][C++0x] std::async fails to compile with simple tests, including N3000 example

[paolo dot carlini at oracle dot com]

------- Comment #31 from paolo dot carlini at oracle dot com  2010-02-12 22:34 -------
Done.


-- 

paolo dot carlini at oracle dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|SUSPENDED                   |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42819