Thread overview
Re: Is there any convenient CopyMemory function in Phobos?
Jun 13, 2011
Jonathan M Davis
Jun 14, 2011
Andrej Mitrovic
Jun 14, 2011
bearophile
Jun 14, 2011
Jonathan M Davis
Jun 14, 2011
Andrej Mitrovic
Jun 14, 2011
Jonathan M Davis
Jun 14, 2011
Johannes Pfau
Jun 14, 2011
Andrej Mitrovic
June 13, 2011
On 2011-06-13 16:27, Andrej Mitrovic wrote:
> Apparently in the Windows API there's a whole lot of byte-copying going around.
> 
> Here's an example:
> 
> int CALLBACK EnhMetaFileProc (HDC hdc, HANDLETABLE * pHandleTable,
> CONST ENHMETARECORD * pEmfRecord,
> int iHandles, LPARAM pData)
> {
> ENHMETARECORD * pEmfr ;
> 
> pEmfr = (ENHMETARECORD *) malloc (pEmfRecord->nSize) ;
> 
> CopyMemory (pEmfr, pEmfRecord, pEmfRecord->nSize) ;
> 
> if (pEmfr->iType == EMR_RECTANGLE)
> pEmfr->iType = EMR_ELLIPSE ;
> 
> PlayEnhMetaFileRecord (hdc, pHandleTable, pEmfr, iHandles) ;
> 
> free (pEmfr) ;
> 
> return TRUE ;
> }
> 
> The nSize field specifies the byte size of the structure pointed to by
> pEmfRecord. The reason copying is required is because of this code:
> if (pEmfr->iType == EMR_RECTANGLE)
> pEmfr->iType = EMR_ELLIPSE ;
> 
> pEmfRecord itself is const, so new memory has to be allocated first and then the memory from pEmfRecord is copied there, and then the iType field is changed. A simple pointer dereference won't copy all the bytes, since this is one of those variable-length structure types you often see in C code.
> 
> In D I've used the slice trick to assign a byte range of data from one
> memory location to another:
> extern (Windows)
> int EnhMetaFileProc(HDC hdc, HANDLETABLE* pHandleTable,
> const ENHMETARECORD* pEmfRecord,
> int iHandles, LPARAM pData)
> {
> ENHMETARECORD* pEmfr;
> immutable newlength = pEmfRecord.nSize;
> 
> pEmfr = cast(ENHMETARECORD*)GC.malloc(newlength);
> 
> (cast(ubyte*)pEmfr)[0..newlength] =
> (cast(ubyte*)pEmfRecord)[0..newlength];
> 
> if (pEmfr.iType == EMR_RECTANGLE)
> pEmfr.iType = EMR_ELLIPSE;
> 
> PlayEnhMetaFileRecord(hdc, pHandleTable, pEmfr, iHandles);
> 
> GC.free(pEmfr);
> return TRUE;
> }
> 
> That's quite ugly.. Is there a Phobos/Druntime and maybe
> platform-independent function that can do byte-memory copies in a
> simple way? Something like this is what I need:
> CopyBytes(dest, src, bytecount);
> 
> Btw, CopyMemory is a WinAPI specific function, but I can't even link to it (it's actually an alias to RtlCopyMemory, but it's not in kernel32.lib for some reason even though MSDN says it is).

mempcy? The C memory functions are all available in D.

- Jonathan M Davis
June 14, 2011
Right, haven't thought about the C functions at all (silly me).
Thanks, this did the trick:
extern (C) void* memcpy(void*, const void*, size_t);
June 14, 2011
On 2011-06-13 17:01, Andrej Mitrovic wrote:
> Right, haven't thought about the C functions at all (silly me).
> Thanks, this did the trick:
> extern (C) void* memcpy(void*, const void*, size_t);

core.stdc.string has memcpy's declaration in it. You should be able to just import it rather than declaring memcpy yourself.

- Jonathan M Davis
June 14, 2011
So why's it in core.stdc.string, instead of say.. core.memory?

Btw, I've had some imports already in the module and it seems these two conflict:

import core.thread;
import core.stdc.string;

void main()
{
    int* p, x;
    memcpy(p, x, 1);
}

test.d(9): Error: core.stdc.string.memcpy at
D:\DMD\dmd2\windows\bin\..\..\src\druntime\import\core\stdc\string.di(11)
conflicts with core.thread.memcpy at
D:\DMD\dmd2\windows\bin\..\..\src\druntime\import\core\thread.di(37)
June 14, 2011
Andrej Mitrovic:

> Right, haven't thought about the C functions at all (silly me).
> Thanks, this did the trick:
> extern (C) void* memcpy(void*, const void*, size_t);

Maybe you are able to wrap that memcpy in a templated D function (named copyMemory) that does what you want in a bit safer way, that contains some static and runtime asserts, etc.

Bye,
bearophile
June 14, 2011
On 2011-06-13 17:38, Andrej Mitrovic wrote:
> So why's it in core.stdc.string, instead of say.. core.memory?
> 
> Btw, I've had some imports already in the module and it seems these two conflict:
> 
> import core.thread;
> import core.stdc.string;
> 
> void main()
> {
>     int* p, x;
>     memcpy(p, x, 1);
> }
> 
> test.d(9): Error: core.stdc.string.memcpy at
> D:\DMD\dmd2\windows\bin\..\..\src\druntime\import\core\stdc\string.di(11)
> conflicts with core.thread.memcpy at
> D:\DMD\dmd2\windows\bin\..\..\src\druntime\import\core\thread.di(37)

core.thread has its own private declaration of memcpy (IIRC, they don't match for some reason). Why a private declaration conflicts, I don't know. That certainly _sounds_ like a bug. Regardless, I have no clue why those declarations are where they are. I would have expected something like core.memory, but they seem to be strewn about druntime. That should probably be improved. I just grepped druntime to find where memcpy's declaration was. I don't know why it is where it is.

- Jonathan M Davis
June 14, 2011
Jonathan M Davis wrote:
>On 2011-06-13 17:38, Andrej Mitrovic wrote:
>> So why's it in core.stdc.string, instead of say.. core.memory?
>> 
>> Btw, I've had some imports already in the module and it seems these two conflict:
>> 
>> import core.thread;
>> import core.stdc.string;
>> 
>> void main()
>> {
>>     int* p, x;
>>     memcpy(p, x, 1);
>> }
>> 
>> test.d(9): Error: core.stdc.string.memcpy at
>> D:\DMD\dmd2\windows\bin\..\..\src\druntime\import\core\stdc\string.di(11)
>> conflicts with core.thread.memcpy at
>> D:\DMD\dmd2\windows\bin\..\..\src\druntime\import\core\thread.di(37)
>
>core.thread has its own private declaration of memcpy (IIRC, they don't match for some reason). Why a private declaration conflicts, I don't know. That certainly _sounds_ like a bug. Regardless, I have no clue why those declarations are where they are. I would have expected something like core.memory, but they seem to be strewn about druntime. That should probably be improved. I just grepped druntime to find where memcpy's declaration was. I don't know why it is where it is.
>
>- Jonathan M Davis

The druntime C bindings follow the names of the C header files. memcpy is in 'string.h' in C --> 'core.stdc.string' in D.

-- 
Johannes Pfau

June 14, 2011
On 6/14/11, Johannes Pfau <spam@example.com> wrote:
> The druntime C bindings follow the names of the C
> header files. memcpy is in 'string.h' in C --> 'core.stdc.string' in D.

I assume the reason for this was because C requires strcpy and memcpy all the time when copying strings?