[Vm-dev] Re: Compiling Cog (merge with eem.87)

[Eliot Miranda]

On Tue, Jun 28, 2011 at 5:33 AM, Igor Stasenko <siguctua at gmail.com> wrote:

>
> On 28 June 2011 12:32, Igor Stasenko <siguctua at gmail.com> wrote:
> > /Users/sig/projects/cog/blessed/src/vm/cogit.c:15864: error: lvalue
> > required as left operand of assignment
> > /Users/sig/projects/cog/blessed/src/vm/cogit.c:15864: error: lvalue
> > required as left operand of assignment
> >
> > here the problematic line:
> >
> >        case SSConstant:
> >                                inst =
> (shouldAnnotateObjectReference((self_in_storeToReg->constant))
> >                        ?
> annotateobjRef(gMoveCwR((self_in_storeToReg->constant), reg),
> > (self_in_storeToReg->constant))
> >>>>                     : /* begin MoveCq:R: */quickConstant =
> (self_in_storeToReg->constant),genoperandoperand(MoveCqR, quickConstant,
> reg));
> >                break;
> >
> >
> > seems like code generator producing bad code.
> > Or maybe i'm not initializing stuff properly (since there are new
> > options which may need to be passed?)
> >
>
> The fix is to change the generator to produce a following:
>
> ({ statement a; statement b; })
>

No. ({ ... }) is a gcc-speciofic extension.  See previous message for
correct fix (checking in now).


>
> instead of
>
> statement a, statement b
>
> If you consider a slang code like:
>
> v := self something ifTrue: [ 1 ] ifFalse: [ x := self doSomething.
> self returnValue].
>
> a generator should produce something like following:
>
> v = something() ? 1 : ({ x = doSomething(); returnValue(); });
>
> Then it will work well (at least in GCC).
>
> I changed the C code to check if compiler can swallow it:
>
> inst = (shouldAnnotateObjectReference((self_in_storeToReg->constant))
>                        ? annotateobjRef(gMoveCwR (
> (self_in_storeToReg->constant), reg),
> (self_in_storeToReg->constant))
>                                                : /* begin MoveCq:R: */({
> quickConstant =
> (self_in_storeToReg->constant); genoperandoperand(MoveCqR,
> quickConstant, reg);}) );
>
>
> Anyways, a code generator needs to be fixed, or code in #storeToReg:
> and #popToReg: needs to be rewritten to not confuse the generator.
>
>
> Another issue is with
> /Users/sig/projects/cog/blessed/src/vm/gcc3x-cointerp.c:10137: error:
> ‘cmDynSuperEntryOffset’ undeclared (first use in this function)
> /Users/sig/projects/cog/blessed/src/vm/gcc3x-cointerp.c:10137: error:
> (Each undeclared identifier is reported only once
> /Users/sig/projects/cog/blessed/src/vm/gcc3x-cointerp.c:10137: error:
> for each function it appears in.)
>
> as far as i see, this variable is used only when compiling newspeak vm
> (there are a lot of places with
> #if NEWSPEAK .. #else , and this variable used only in newspeak blocks.
>
> But somehow it slept through and introduced in cog.
>

Because it is marked as <api> it is retained.  <api> marks methods used
outside the source file (e.g. in cogit.c).  So if it is marked <api> but
optional it needs to be filtered out.  I implemented <option: #SomeKey) for
this and eliminated the <cogit: #CogitClassName> tag, since it is now
subsumed by <option: ...>.

Thanks, Igor!

Probably #ceDynamicSuperSend:to:numArgs:
> should also contain
>        self cppIf: NewspeakVM
>                ifTrue: []
> clause.
>
> > --
> > Best regards,
> > Igor Stasenko AKA sig.
> >
>
>
>
> --
> Best regards,
> Igor Stasenko AKA sig.
>



-- 
best,
Eliot
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20110628/8a6c65ed/attachment.htm