[prev in list] [next in list] [prev in thread] [next in thread] 

List:       wine-devel
Subject:    Re: Make more of the OLE interface vtables const
From:       Marcus Meissner <marcus () jet ! franken ! de>
Date:       2005-05-30 5:02:19
Message-ID: 20050530050219.GA7788 () jet ! franken ! de
[Download RAW message or body]

On Mon, May 30, 2005 at 11:33:28AM +0900, Dmitry Timoshkov wrote:
> "Andreas Mohr" <andi@rhlx01.fht-esslingen.de> wrote:
> 
> > On Sun, May 29, 2005 at 07:05:49PM +0900, Dmitry Timoshkov wrote:
> > > Hello,
> > > 
> > > Changelog:
> > >     Dmitry Timoshkov <dmitry@codeweavers.com>
> > >     Make more of the OLE interface vtables const.
> > You do know that those will *not* end up const?
> > 
> > filemoniker.c:static const IClassFactoryVtbl FileMonikerCFVtbl =
> > objdump -x ole32.dll.so:
> > 0008efb0 l     O .data  00000014              FileMonikerCFVtbl
> > 
> > My suspicion is that this is due to winegcc compile, all other binaries
> > with normal compile (e.g. winedump, ...) have all their const data in .rodata segment...
> > 
> > So, does anybody know what seems to be wrong with winegcc or why it chose to do that?
> 
> It appears to be not a winegcc problem. That's how gcc generates pointer tables
> (regardless of the pointer type). Although it (gcc) correctly produces warnings
> about losing 'const' attribute while passing a pointer to a const table around, it
> still places contents of the table into r/w section. I'd call it a gcc limitation,
> if not a bug.

The problem is that the pointers in the table need relocation, so it needs to
be modified during linking, so it cannot be fully const.

Ciao, Marcus

[prev in list] [next in list] [prev in thread] [next in thread]