[Bug c++/54413] New: Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[Bug c++/54413] New: Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[p4spam at hispeed dot ch]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

             Bug #: 54413
           Summary: Option for turning off compiler extensions for
                    -std=c++11 with respect to complex/fixed-point numbers
                    missing
    Classification: Unclassified
           Product: gcc
           Version: 4.7.1
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: c++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: p4spam@hispeed.ch


All versions of g++ that support UDL operators should have such an option to
turn off GNU complex literals and fixed point literals extensions.

I am prototyping C++1x library extensions for user-defined literals. While it
is possible to turn off GCC's extensions for complex numbers for "old" C++ it
is impossible to combine the effects of -ansi with -std=c++11. Especially the
syntax extensions for complex numbers, e.g., 1.1i cannot be turned off and thus
do not allow testing of UDL operators for the purpose of creating
std::complex<> literals, such as

constexpr std::complex<double> operator"" i(long double d){
return std::complex<double>(0,d);
}

>From the fixed-point extension also number suffixes r and R are occupied.

More info at my proposal, 
http://wiki.hsr.ch/PeterSommerlad/files/N3402_UDLsuffixes.pdf

All versions of g++ that support UDL operators should have such an option to
turn off GNU complex literals and fixed point literals extensions.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-08-30 09:10:58 UTC ---
[usrlit.suffix]/1 says that udlit suffixes that don't start with underscore are
reserved.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-08-30
     Ever Confirmed|0                           |1
           Severity|major                       |normal

--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-08-30 09:55:31 UTC ---
(In reply to comment #0)
> While it
> is possible to turn off GCC's extensions for complex numbers for "old" C++ it

Is it? I don't think that's true.

> is impossible to combine the effects of -ansi with -std=c++11.

That's because -ansi means -std=c++98, it has nothing to do with whether
complex literals are supported.

(In reply to comment #1)
> [usrlit.suffix]/1 says that udlit suffixes that don't start with underscore are
> reserved.

Yes, for future UDLs. Peter's paper proposes new standard UDLs for standard
library types so in order for G++ to support them it must be able to tell the
preprocessor not to treat 1i as a complex literal, so the front-end can turn it
into a call to a UDL operator in the standard library.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-08-30 10:03:51 UTC ---
Perhaps we should for i and r etc. suffixes in C++11+ first try to look it up
as user-defined-literal and only if there are no corresponding operator""s,
treat it as the GNU extension?

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |3dw4rd at verizon dot net

--- Comment #4 from Paolo Carlini <paolo.carlini at oracle dot com> 2012-10-01 16:51:57 UTC ---
Adding Ed in CC.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #5 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-10-01 18:47:33 UTC ---
If I remember correctly, we are supposed to parse compiler-specific extensions
before UDLs - i.e.  The compiler extensions get dibs.  

I guess that's not the issue.  You propose being able to turn off (presumably
all?) gcc-specific suffix extensions.  Presumably, these should be on by
default for C++1*?  I agree we should have something like this.

Detecting at the level of c-family or the lexer whether c++ has seen a
definition of operator"" r or i could be ... real interesting.

Perhaps we could "move" these gnu extensions to:
namespace __gnucxx{
  operator"" r()
}
somehow.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #6 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-10-15 21:08:27 UTC ---
So what should these flags look like?

-fno-complex-literals
-fno-fixed-literals

Or should I use -W?

Should -std=c++11, etc. set these by default 
(and you turn then back on with -fcomplex-literals, -ffixed-literals?)

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[peter.sommerlad at hsr dot ch]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #7 from Peter Sommerlad <peter.sommerlad at hsr dot ch> 2012-10-15 21:17:07 UTC ---
(In reply to comment #6)
> So what should these flags look like?
> 
> -fno-complex-literals
> -fno-fixed-literals
looks fine in principle.
> 
> Or should I use -W?
> 
> Should -std=c++11, etc. set these by default 
> (and you turn then back on with -fcomplex-literals, -ffixed-literals?)

I would happy with -ansi -std=c++11 work as I expected.

turning the extensions off with just -std=c++11 would not be what you want,
since existing code might rely on these extensions, only when we get -std=c++16
with my paper on UDL accepted, we should turn off these extensions, because
then the standard library will ship with the UDL suffixes identical or closely
to the GNU extensions.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #8 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-10-15 21:19:45 UTC ---
(In reply to comment #7)
> I would happy with -ansi -std=c++11 work as I expected.

Again, -ansi is a synonym for -std=c++98, so saying "-ansi -std=c++11" is
completely pointless.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[peter.sommerlad at hsr dot ch]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #9 from Peter Sommerlad <peter.sommerlad at hsr dot ch> 2012-10-15 21:21:17 UTC ---

Then let us just decide on the following two:

-fno-complex-literals
-fno-fixed-literals

> looks fine in principle.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #10 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-10-15 21:25:23 UTC ---
So maybe we can enable by default for -std=c++1y, etc.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #11 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-11-05 04:50:20 UTC ---
Here is a patch that should work.  it passes on x86_64 linux.
I would like to get this in for 4.8 if possible.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #12 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-11-05 04:55:36 UTC ---
Created attachment 28617
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28617
Patch to implement flags allowing gnu suffixes to be used as user-defined
literals.


Here is a patch that should work.  it passes on x86_64 linux.

I would like to get this in for 4.8 if possible.

There are three flags:

-f[no-]imaginary-literals: Frees 'i', 'I', 'j', 'J'
-f[no-]fixed-point-literals: Frees 'k', 'K', 'r', 'R'
-f[no-]machine-defined-literals: Frees 'w', 'W', 'q', 'Q'

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #13 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-11-06 17:54:31 UTC ---
The patch tests clean on x86_64-linux.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

Ed Smith-Rowland <3dw4rd at verizon dot net> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #28617|0                           |1
        is obsolete|                            |

--- Comment #14 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-11-12 05:39:13 UTC ---
Created attachment 28665
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28665
Committed patch.

Committed the following.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #15 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-11-12 09:12:35 UTC ---
Ed, N.B. if you use the ChangeLog entry as the svn commit log then the PR
number in the log means Bugzilla gets automatically updated with a link to the
commit, and you don't need to attach it

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #16 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-11-12 17:48:29 UTC ---
Thanks, So If there are several ChangeLogs in the tree to get updated which one
do I put in the svn commit?  Or does it matter?

Also, I just realized that maybe I should whip up a sentence or two for
http://gcc.gnu.org/gcc-4.8/changes.html?

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #17 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-11-12 18:30:27 UTC ---
(In reply to comment #16)
> Thanks, So If there are several ChangeLogs in the tree to get updated which one
> do I put in the svn commit?  Or does it matter?

I put all of them, prefixed by the dir e.g.
http://gcc.gnu.org/viewcvs?view=revision&revision=192968

> Also, I just realized that maybe I should whip up a sentence or two for
> http://gcc.gnu.org/gcc-4.8/changes.html?

That would be great, thanks.

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #18 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-11-14 13:07:00 UTC ---
I added a bullet for this flag in gcc-4.8/changes.html.

How does one close a bug?

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.8.0

--- Comment #19 from Paolo Carlini <paolo.carlini at oracle dot com> 2012-11-14 13:18:18 UTC ---
Doesn't seem so ambiguous to me ;)

[Bug c++/54413] Option for turning off compiler extensions for -std=c++11 with respect to complex/fixed-point numbers missing

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54413

--- Comment #20 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-11-14 13:21:06 UTC ---
(In reply to comment #18)
> I added a bullet for this flag in gcc-4.8/changes.html.

Thanks!

> How does one close a bug?

You need to have the right permission bits set on your account, which are
automatically set if you use an @gcc.gnu.org account.