Thread: BUG #13788: compile error in generic_msvc.h

BUG #13788: compile error in generic_msvc.h

From
paul.moore@centrify.com
Date:
The following bug has been logged on the website:

Bug reference:      13788
Logged by:          paul moore
Email address:      paul.moore@centrify.com
PostgreSQL version: 9.5beta1
Operating system:   win server 2012
Description:

vs 2015, compiling spi execq sample

vc complains about this

1>c:\Program
Files\PostgreSQL\9.5\include\server\port/atomics/generic-msvc.h(91): error
C2664: 'LONG64 _InterlockedCompareExchange64(volatile LONG64
*,LONG64,LONG64)': cannot convert argument 1 from 'volatile uint64 *' to
'volatile LONG64 *'

same for the add64 function a few lines later

Re: BUG #13788: compile error in generic_msvc.h

From
Tom Lane
Date:
paul.moore@centrify.com writes:
> Bug reference:      13788
> Logged by:          paul moore
> Email address:      paul.moore@centrify.com
> PostgreSQL version: 9.5beta1
> Operating system:   win server 2012
> Description:

> vs 2015, compiling spi execq sample

> vc complains about this

> 1>c:\Program
> Files\PostgreSQL\9.5\include\server\port/atomics/generic-msvc.h(91): error
> C2664: 'LONG64 _InterlockedCompareExchange64(volatile LONG64
> *,LONG64,LONG64)': cannot convert argument 1 from 'volatile uint64 *' to
> 'volatile LONG64 *'

> same for the add64 function a few lines later

That's a bit weird.  A perhaps plausible explanation would be that the
compiler is unhappy about passing an unsigned argument to a signed
parameter; but if they've tightened it up like that, why no similar
complaint for the 32-bit versions just above these?

Can you try inserting a cast to (volatile int64 *), just to see?

            regards, tom lane

Re: BUG #13788: compile error in generic_msvc.h

From
Andres Freund
Date:
On 2015-12-02 16:13:07 +0000, Paul Moore wrote:
> The 32 bit functions are defined on unsigned types
>
> From winbase.h
>
> FORCEINLINE
> unsigned
> InterlockedCompareExchange(
>     _Inout_ _Interlocked_operand_ unsigned volatile *Destination,
>     _In_ unsigned Exchange,
>     _In_ unsigned Comperand
>     )
> {
>     return (unsigned) _InterlockedCompareExchange((volatile long*) Destination, (long) Exchange, (long) Comperand);
> }
>
> The cast fixes the 64 bit case.
>
> Full disclosure, I am compiling as c++. I think that might be the cause. I tried in a c project and did not get the
sameerror 

Given that win64 buildfarm members have happily been compiling this code
I suspect that's the cause. Are you including the headers with extern
"C" or not?

You're using SPI directly from C++? You got to be *very* careful doing
that - postgres uses longjmp which isn't necessarily compatible with C++
(doesn't call destructors to be called and such).

Re: BUG #13788: compile error in generic_msvc.h

From
Paul Moore
Date:
The 32 bit functions are defined on unsigned types

From winbase.h

FORCEINLINE
unsigned
InterlockedCompareExchange(
    _Inout_ _Interlocked_operand_ unsigned volatile *Destination,
    _In_ unsigned Exchange,
    _In_ unsigned Comperand
    )
{
    return (unsigned) _InterlockedCompareExchange((volatile long*) Destinat=
ion, (long) Exchange, (long) Comperand);
}

The cast fixes the 64 bit case.

Full disclosure, I am compiling as c++. I think that might be the cause. I =
tried in a c project and did not get the same error

-----Original Message-----
From: Tom Lane [mailto:tgl@sss.pgh.pa.us]=20
Sent: Tuesday, December 1, 2015 1:35 PM
To: Paul Moore <paul.moore@centrify.com>
Cc: pgsql-bugs@postgresql.org; Andres Freund <andres@anarazel.de>
Subject: Re: [BUGS] BUG #13788: compile error in generic_msvc.h

paul.moore@centrify.com writes:
> Bug reference:      13788
> Logged by:          paul moore
> Email address:      paul.moore@centrify.com
> PostgreSQL version: 9.5beta1
> Operating system:   win server 2012
> Description:       =20

> vs 2015, compiling spi execq sample

> vc complains about this

> 1>c:\Program
> Files\PostgreSQL\9.5\include\server\port/atomics/generic-msvc.h(91):=20
> error
> C2664: 'LONG64 _InterlockedCompareExchange64(volatile LONG64
> *,LONG64,LONG64)': cannot convert argument 1 from 'volatile uint64 *'=20
> to 'volatile LONG64 *'

> same for the add64 function a few lines later

That's a bit weird.  A perhaps plausible explanation would be that the comp=
iler is unhappy about passing an unsigned argument to a signed parameter; b=
ut if they've tightened it up like that, why no similar complaint for the 3=
2-bit versions just above these?

Can you try inserting a cast to (volatile int64 *), just to see?

            regards, tom lane

Re: BUG #13788: compile error in generic_msvc.h

From
Paul Moore
Date:
Yes on extern "C". I have other stuff working

Really what I am trying to do is invoke c# code. I am going to need to call=
 SPI. I tried to find out if anybody else had done it and found nothing.

The c++ code is simply a thin shim from native c to c#

-----Original Message-----
From: Andres Freund [mailto:andres@anarazel.de]=20
Sent: Wednesday, December 2, 2015 8:50 AM
To: Paul Moore <paul.moore@centrify.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>; pgsql-bugs@postgresql.org
Subject: Re: [BUGS] BUG #13788: compile error in generic_msvc.h

On 2015-12-02 16:13:07 +0000, Paul Moore wrote:
> The 32 bit functions are defined on unsigned types
>=20
> From winbase.h
>=20
> FORCEINLINE
> unsigned
> InterlockedCompareExchange(
>     _Inout_ _Interlocked_operand_ unsigned volatile *Destination,
>     _In_ unsigned Exchange,
>     _In_ unsigned Comperand
>     )
> {
>     return (unsigned) _InterlockedCompareExchange((volatile long*)=20
> Destination, (long) Exchange, (long) Comperand); }
>=20
> The cast fixes the 64 bit case.
>=20
> Full disclosure, I am compiling as c++. I think that might be the=20
> cause. I tried in a c project and did not get the same error

Given that win64 buildfarm members have happily been compiling this code I =
suspect that's the cause. Are you including the headers with extern "C" or =
not?

You're using SPI directly from C++? You got to be *very* careful doing that=
 - postgres uses longjmp which isn't necessarily compatible with C++ (doesn=
't call destructors to be called and such).

Re: BUG #13788: compile error in generic_msvc.h

From
Andres Freund
Date:
Hi Paul,

As I don't have access to msvc: Could you try what Tom suggested in
http://archives.postgresql.org/message-id/5176.1449005701%40sss.pgh.pa.us ?

Regards,

Andres

On 2015-12-03 17:02:31 +0000, Paul Moore wrote:
> Yes on extern "C". I have other stuff working
>
> Really what I am trying to do is invoke c# code. I am going to need to call SPI. I tried to find out if anybody else
haddone it and found nothing. 
>
> The c++ code is simply a thin shim from native c to c#
>
> -----Original Message-----
> From: Andres Freund [mailto:andres@anarazel.de]
> Sent: Wednesday, December 2, 2015 8:50 AM
> To: Paul Moore <paul.moore@centrify.com>
> Cc: Tom Lane <tgl@sss.pgh.pa.us>; pgsql-bugs@postgresql.org
> Subject: Re: [BUGS] BUG #13788: compile error in generic_msvc.h
>
> On 2015-12-02 16:13:07 +0000, Paul Moore wrote:
> > The 32 bit functions are defined on unsigned types
> >
> > From winbase.h
> >
> > FORCEINLINE
> > unsigned
> > InterlockedCompareExchange(
> >     _Inout_ _Interlocked_operand_ unsigned volatile *Destination,
> >     _In_ unsigned Exchange,
> >     _In_ unsigned Comperand
> >     )
> > {
> >     return (unsigned) _InterlockedCompareExchange((volatile long*)
> > Destination, (long) Exchange, (long) Comperand); }
> >
> > The cast fixes the 64 bit case.
> >
> > Full disclosure, I am compiling as c++. I think that might be the
> > cause. I tried in a c project and did not get the same error
>
> Given that win64 buildfarm members have happily been compiling this code I suspect that's the cause. Are you
includingthe headers with extern "C" or not? 
>
> You're using SPI directly from C++? You got to be *very* careful doing that - postgres uses longjmp which isn't
necessarilycompatible with C++ (doesn't call destructors to be called and such). 
>
>
> --
> Sent via pgsql-bugs mailing list (pgsql-bugs@postgresql.org)
> To make changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-bugs
Greetings,

Andres Freund

Re: BUG #13788: compile error in generic_msvc.h

From
Paul Moore
Date:
I put the cast in and it compiles with no complaint

-----Original Message-----
From: Andres Freund [mailto:andres@anarazel.de]=20
Sent: Thursday, December 10, 2015 7:14 AM
To: Paul Moore <paul.moore@centrify.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>; pgsql-bugs@postgresql.org
Subject: Re: [BUGS] BUG #13788: compile error in generic_msvc.h

Hi Paul,

As I don't have access to msvc: Could you try what Tom suggested in http://=
archives.postgresql.org/message-id/5176.1449005701%40sss.pgh.pa.us ?

Regards,

Andres

On 2015-12-03 17:02:31 +0000, Paul Moore wrote:
> Yes on extern "C". I have other stuff working
>=20
> Really what I am trying to do is invoke c# code. I am going to need to ca=
ll SPI. I tried to find out if anybody else had done it and found nothing.
>=20
> The c++ code is simply a thin shim from native c to c#
>=20
> -----Original Message-----
> From: Andres Freund [mailto:andres@anarazel.de]
> Sent: Wednesday, December 2, 2015 8:50 AM
> To: Paul Moore <paul.moore@centrify.com>
> Cc: Tom Lane <tgl@sss.pgh.pa.us>; pgsql-bugs@postgresql.org
> Subject: Re: [BUGS] BUG #13788: compile error in generic_msvc.h
>=20
> On 2015-12-02 16:13:07 +0000, Paul Moore wrote:
> > The 32 bit functions are defined on unsigned types
> >=20
> > From winbase.h
> >=20
> > FORCEINLINE
> > unsigned
> > InterlockedCompareExchange(
> >     _Inout_ _Interlocked_operand_ unsigned volatile *Destination,
> >     _In_ unsigned Exchange,
> >     _In_ unsigned Comperand
> >     )
> > {
> >     return (unsigned) _InterlockedCompareExchange((volatile long*)=20
> > Destination, (long) Exchange, (long) Comperand); }
> >=20
> > The cast fixes the 64 bit case.
> >=20
> > Full disclosure, I am compiling as c++. I think that might be the=20
> > cause. I tried in a c project and did not get the same error
>=20
> Given that win64 buildfarm members have happily been compiling this code =
I suspect that's the cause. Are you including the headers with extern "C" o=
r not?
>=20
> You're using SPI directly from C++? You got to be *very* careful doing th=
at - postgres uses longjmp which isn't necessarily compatible with C++ (doe=
sn't call destructors to be called and such).
>=20
>=20
> --
> Sent via pgsql-bugs mailing list (pgsql-bugs@postgresql.org) To make=20
> changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-bugs
Greetings,

Andres Freund