Date: Tue, 10 Sep 2002 22:37:47 +0200 From: Bernd Walter <ticso@cicely5.cicely.de> To: Andrew Gallatin <gallatin@cs.duke.edu>, Fred Clift <fclift@verio.net> Cc: ticso@cicely.de, Joe Marcus Clarke <marcus@marcuscom.com>, alpha@FreeBSD.ORG Subject: Re: getting closer (was Re: mozilla failure on 4.6.2) Message-ID: <20020910203747.GE399@cicely5.cicely.de> In-Reply-To: <20020910112155.E32234-100000@vespa.dmz.orem.verio.net> <15742.812.130498.758611@grasshopper.cs.duke.edu> References: <15737.2550.488154.4401@grasshopper.cs.duke.edu> <20020907000825.GA4091@dragon.nuxi.com> <15739.43883.152090.936327@grasshopper.cs.duke.edu> <15741.2822.526047.253409@grasshopper.cs.duke.edu> <1031605410.351.90.camel@gyros.marcuscom.com> <15741.17468.701009.998323@grasshopper.cs.duke.edu> <1031635953.342.133.camel@gyros.marcuscom.com> <15741.61382.872535.426732@grasshopper.cs.duke.edu> <20020910140414.GC5057@cicely5.cicely.de> <15742.812.130498.758611@grasshopper.cs.duke.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Sep 10, 2002 at 10:35:24AM -0400, Andrew Gallatin wrote:
>
> Bernd Walter writes:
> > On Tue, Sep 10, 2002 at 09:12:38AM -0400, Andrew Gallatin wrote:
> > >
> > > Joe Marcus Clarke writes:
> > > > Add 8 to the stack pointer ("this", right), and store it in $16.
> > >
> > > I wish it were that simple :-(
> > >
> > > The problem seems to be that the application code is passing the
> > > "wrong" value in and somehow the c++ calling standard corrects it, and
> > > we must correct it too:
> > >
> > > impl == 0x12001c080
> > > foo == 0x12001c080
> > > bar == 0x12001c088
> >
> > I have no idea how the class declaration for you example looks like.
>
> http://lxr.mozilla.org/seamonkey/source/xpcom/reflect/xptcall/tests/TestXPTCInvoke.cpp
>
> > Possible because I don't know the mozilla code you are talking about.
> > What is the difference in direct and invoke calling?
>
> "direct" is a normal c++ call: bar->BarMethod1(1);
>
>
> "invoke" means the mozilla xptinvoke shims:
>
> var[0].val.i32 = 1;
> var[0].type = nsXPTType::T_I32;
> var[0].flags = 0;
> XPTC_InvokeByIndex(bar, 3, 1, var);
>
>
> They are described at:
> http://www.mozilla.org/scriptable/xptcall-faq.html
OK - I see.
Well that's evil - I had wished they would at least print a clear
warning so everyone would know that specific platform support is
missing instead of just producing nameless build errors.
> > Lets asume foo is a virtual function of class x which is based on
> > class y defining the virtual function bar.
> > >From what I know about C++ (which might be outdated) we have the
> > following situation when using an object of class x.
> > Call of function bar means the caller knows the offset to the function
> > pointer, because it's using a pointer of type x.
> > The function bar expects a this pointer of type y so it gets the
> > this pointer of the embedded y object which may be numericaly different.
> >
> > Now asume class x reimplement the virtual function bar.
> > The newly bar function has to share the calling conventions that were
> > declared by class y.
> > If we call function bar using a type y pointer we submit a this pointer
> > of type y.
> > But the function is designed to have a this pointer of type x.
> > Now the function must recalculate the supplied y pointer into an x
> > pointer.
> > If we call funtcion bar using a type x pointer we still have to convert
> > the x pointer into an y pointer because the function still expects an y
> > pointer.
>
> I assume this is all done "behind the scenes" by the compiler?
Normaly yes.
In that case they do it theirself.
> Is there a convention for where the conversions are passed?
No clue - I'm just a c++ user - not a compiler writer...
My knowledge about vtables is just basic to understand how virtual
functions could influence performance.
On Tue, Sep 10, 2002 at 11:24:04AM -0600, Fred Clift wrote:
> On Tue, 10 Sep 2002, Andrew Gallatin wrote:
>
> >
> > Since you know quite a bit about c++, and because you know alpha asm,
> > I'd _really_ like your help ;) Do you have an alpha running
> > -stable? Or are you all -current these days?
>
> My knowlege of both c++ and alpha asm is very lacking, but, I have a PWS
> 433au running a fairly recent -stable (jul 20 - I could upgrade) on that I
> could be pursuaded to set up a useraccount or 3 on if it would help with
> this problem.
Fred - thank you for that offer but I have access to a stable alpha.
cvsup.de.freebsd.org is my machine and is running stable from 31th jul.
My home alphas are both running -current.
I can boot stable on my NoName - but compiling mozilla - better not...
My problem is mostly time.
E.g. I recently added an USB controller to my NoName and I promissed
Jonathan Mini alpha MD part for his KSE work.
--
B.Walter COSMO-Project http://www.cosmo-project.de
ticso@cicely.de Usergroup info@cosmo-project.de
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-alpha" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20020910203747.GE399>