llvm dev - Jan 2013 - [LLVMdev] Using C++'11 language features in LLVM itself

On Sun, Jan 13, 2013 at 10:26 AM, Tim Northover <t.p.northover at
gmail.com> wrote:
>> Good points David, I don't feel qualified to evaluate the
differences
>> between those versions though... Perhaps Richard or Doug could comment
here?
>
> Unless I'm misreading the buildbots, we don't actually have
anything
> trying to build with MSVC. Have we considered how we're going to
> enforce these rather subtle distinctions?
The buildbots on lab.llvm.org don't cover Windows, but Takumi
maintains Windows buildbots & fairly diligently (autoreporting to IRC,
plus manual follow up on review threads, etc) maintains them. Some
other developers (such as Michael Spencer) work on Windows on a
regular basis, so they'll essentially come screaming if it breaks ;)

On Sun, Jan 13, 2013 at 7:39 PM, David Blaikie <dblaikie at gmail.com>
wrote:
> On Sun, Jan 13, 2013 at 10:26 AM, Tim Northover <t.p.northover at
gmail.com>
> wrote:
> >> Good points David, I don't feel qualified to evaluate the
differences
> >> between those versions though... Perhaps Richard or Doug could
comment
> here?
> >
> > Unless I'm misreading the buildbots, we don't actually have
anything
> > trying to build with MSVC. Have we considered how we're going to
> > enforce these rather subtle distinctions?
>
> The buildbots on lab.llvm.org don't cover Windows, but Takumi
> maintains Windows buildbots & fairly diligently (autoreporting to IRC,
> plus manual follow up on review threads, etc) maintains them. Some
> other developers (such as Michael Spencer) work on Windows on a
> regular basis, so they'll essentially come screaming if it breaks ;)
>

Following the feedback from David, here is an updated list of the profiles,
taking into account deprecated versions of the specifications of the
lambdas and r-values that might be of interest. As before, they are duly
annotated with a remark.

gcc 4.3, MSVC 10, clang 3.0
=> main benefits are decltype v1.0 [1], static_assert and built-in type
traits.

gcc 4.4, MSVC 10, clang 3.0
=> main additions: late specified return type.

gcc 4.5, MSVC 10, clang 3.1
=> main additions: lambda v1.0 [2], local types as template arguments,
r-value v2.0 [3].

gcc 4.5, MSVC 11, clang 3.1
=> main additions: atomic, lambda v1.1, r-value v2.1 [3], strongly typed
enum.

gcc 4.6, MSVC 11, clang 3.1
=> main additions: forward enum, nullptr, range-based for.

gcc 4.7, MSVC 11, clang 3.1
=> main additions: override/final, template aliases.

gcc 4.7, MSVC 11 nov' 12, clang 3.1
=> main additions: initializer lists, variadic templates v1.0 [4].

Note: the only benefit of gcc 4.6 + MSVC10 over gcc 4.5 + MSVC10 is
"nullptr".


As we can see, if we are confident enough in the Windows buildbots to pick
up the differences between the various versions (and I suppose we already
rely on them), then the version proposed by Chris does provide a number of
new C++11 features compared to the current state:

gcc 4.5, MSVC 10, clang 3.1
- decltype v1.0 [1] + late specified return type
- lambda v1.0 [2]
- local types as template arguments
- r-value 2.0 [3]
- static_assert
- built-in type traits

The only issue I can see is that a number of people emitted the idea that
gcc 4.5 might be too advanced for a number of platforms (including RedHat
for example).

-- Matthieu


[1]: "decltype v1.0 (n2343)" is subsided by "decltype v1.1
(n3276)" which
drops the completeness requirement of return types; unfortunately it seems
no released version of gcc implements v1.1.

[2]: "lambda v1.0 (n2658)" is subsided by "lambda v1.1
(n2927)", not sure
of the exact impact of the changes.

[3]: "r-value v2.0 (n2844)" and "r-value v2.1 (n2844 + core issue
1138)"
are subsided by "r-value v3.0 (n3053)" once again not sure of the
impact of
the changes; furthermore it seems no released version of MSVC implements
v3.0.

[4]: a prior implementation of variadic templates (dubbed v0.9) was
available as soon as gcc 4.4 (with issues regarding template template
parameters matching), however MSVC never implemented it and jumped straight
to v1.0 in the CTP release.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130113/10e1bb1b/attachment.html>

On Sun, Jan 13, 2013 at 2:10 PM, Matthieu Monrocq <
matthieu.monrocq at gmail.com> wrote:
>
>
> On Sun, Jan 13, 2013 at 7:39 PM, David Blaikie <dblaikie at
gmail.com> wrote:
>
>> On Sun, Jan 13, 2013 at 10:26 AM, Tim Northover <t.p.northover at
gmail.com>
>> wrote:
>> >> Good points David, I don't feel qualified to evaluate the
differences
>> >> between those versions though... Perhaps Richard or Doug could
comment
>> here?
>> >
>> > Unless I'm misreading the buildbots, we don't actually
have anything
>> > trying to build with MSVC. Have we considered how we're going
to
>> > enforce these rather subtle distinctions?
>>
>> The buildbots on lab.llvm.org don't cover Windows, but Takumi
>> maintains Windows buildbots & fairly diligently (autoreporting to
IRC,
>> plus manual follow up on review threads, etc) maintains them. Some
>> other developers (such as Michael Spencer) work on Windows on a
>> regular basis, so they'll essentially come screaming if it breaks
;)
>>
>
>
> Following the feedback from David, here is an updated list of the
> profiles, taking into account deprecated versions of the specifications of
> the lambdas and r-values that might be of interest. As before, they are
> duly annotated with a remark.
>
>
> gcc 4.3, MSVC 10, clang 3.0
> => main benefits are decltype v1.0 [1], static_assert and built-in type
> traits.
>
>
> gcc 4.4, MSVC 10, clang 3.0
> => main additions: late specified return type.
>
> gcc 4.5, MSVC 10, clang 3.1
> => main additions: lambda v1.0 [2], local types as template arguments,
> r-value v2.0 [3].
>
>
> gcc 4.5, MSVC 11, clang 3.1
> => main additions: atomic, lambda v1.1, r-value v2.1 [3], strongly typed
> enum.
>
> gcc 4.6, MSVC 11, clang 3.1
> => main additions: forward enum, nullptr, range-based for.
>
> gcc 4.7, MSVC 11, clang 3.1
> => main additions: override/final, template aliases.
>
> gcc 4.7, MSVC 11 nov' 12, clang 3.1
> => main additions: initializer lists, variadic templates v1.0 [4].
>
> Note: the only benefit of gcc 4.6 + MSVC10 over gcc 4.5 + MSVC10 is
> "nullptr".
>
>
> As we can see, if we are confident enough in the Windows buildbots to pick
> up the differences between the various versions (and I suppose we already
> rely on them), then the version proposed by Chris does provide a number of
> new C++11 features compared to the current state:
>
> gcc 4.5, MSVC 10, clang 3.1
> - decltype v1.0 [1] + late specified return type
> - lambda v1.0 [2]
> - local types as template arguments
> - r-value 2.0 [3]
> - static_assert
> - built-in type traits
>
> The only issue I can see is that a number of people emitted the idea that
> gcc 4.5 might be too advanced for a number of platforms (including RedHat
> for example).
>
The current crop of enterprise linux distros are on 4.4.  Getting a clang
3.2 binary on these systems isn't difficult, though.

>
> -- Matthieu
>
>
> [1]: "decltype v1.0 (n2343)" is subsided by "decltype v1.1
(n3276)" which
> drops the completeness requirement of return types; unfortunately it seems
> no released version of gcc implements v1.1.
>
> [2]: "lambda v1.0 (n2658)" is subsided by "lambda v1.1
(n2927)", not sure
> of the exact impact of the changes.
>
> [3]: "r-value v2.0 (n2844)" and "r-value v2.1 (n2844 + core
issue 1138)"
> are subsided by "r-value v3.0 (n3053)" once again not sure of the
impact of
> the changes; furthermore it seems no released version of MSVC implements
> v3.0.
>
>
> [4]: a prior implementation of variadic templates (dubbed v0.9) was
> available as soon as gcc 4.4 (with issues regarding template template
> parameters matching), however MSVC never implemented it and jumped straight
> to v1.0 in the CTP release.
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
>

-- 

Thanks,

Justin Holewinski
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130113/c3e4529c/attachment.html>

On Sun, Jan 13, 2013 at 2:10 PM, Matthieu Monrocq
<matthieu.monrocq at gmail.com> wrote:
> gcc 4.5, MSVC 10, clang 3.1
> - decltype v1.0 [1] + late specified return type
> - lambda v1.0 [2]
> - local types as template arguments
> - r-value 2.0 [3]
> - static_assert
> - built-in type traits
This isn't very encouraging. Anecdotally from what I've seen in LLD
(whose C++11 feature use is bounded only by MSVC 2012), the features
that get the most mileage by a large margin are `auto`, range-for, and
`nullptr`, along with library offerings like unique_ptr<>. Just based
on that, the churn and overhead of imposing new build requirements
doesn't seem worth it at the moment.

-- Sean Silva

Matthieu Monrocq <matthieu.monrocq at gmail.com> writes:
> [2]: "lambda v1.0 (n2658)" is subsided by "lambda v1.1
(n2927)", not sure
> of the exact impact of the changes.
>
> [3]: "r-value v2.0 (n2844)" and "r-value v2.1 (n2844 + core
issue 1138)"
> are subsided by "r-value v3.0 (n3053)" once again not sure of the
impact of
> the changes; furthermore it seems no released version of MSVC implements
> v3.0.
Details on supported features by VS are here:

http://blogs.msdn.com/b/vcblog/archive/2011/09/12/10209291.aspx

The document also contains clarifications about the different revisions
of the specification for some features.

> Following the feedback from David, here is an updated list of the profiles,
> taking into account deprecated versions of the specifications of the
lambdas
> and r-values that might be of interest. As before, they are duly annotated
> with a remark.
It's much smaller than these headline features, but how about commas
at the end of enum lists? Could be useful for keeping revision-control
logs in good shape.

Tim.