[M3devel] Static chains

[Tony Hosking]

Yes, I am leery myself of much the same.  Gcc is very careful about  
when a nested procedure is allocated a static link.

On Jan 15, 2008, at 12:54 AM, Jay wrote:

> I really don't think an optimizing compiler should be passing  
> around unused static links just for the debugger to use. The  
> debugger should look up the stack itself.
>
> Optimizing compilers remove locals, remove parameters, remove  
> functions, create custom calling conventions, etc.
>
> An non-optimizing compiler, sure, I guess.
>
> Though it seems some level of optimization should be always on.
> It pains me to think that every little function call I write might  
> be made, and not inlined.
> Microsoft compilers now do inlining across static lib boundaries.
> Or that constant arithmetic like:
> #define A 1
> #define B 2
> int c = (A + B)
>
> might be computed at runtime.
>
> Or that
> while(1)
>   ..
>   .. if ... break;
>
> might generate a runtime check if 1 is not zero (along with the  
> code bloat).
>
> Microsoft has "__forceinline", and it is a warning or error if the  
> compiler is unable to inline such a function.
>
>  - Jay
>
>
> > Date: Mon, 14 Jan 2008 17:24:35 -0600
> > From: rodney.bates at wichita.edu
> > To: hosking at cs.purdue.edu
> > CC: m3devel at elegosoft.com
> > Subject: Re: [M3devel] Static chains
> >
> > Think of the debugger as an interpreter for a substantial subset  
> of the
> > language. When the program is stopped at some point, m3gdb can,  
> in effect,
> > inject code to evaluate an arbitrary expression, typed by the  
> debugger
> > user, or execute an arbitrary assignment or call statement. Although
> > there are a number of gaps yet in the implementation, m3gdb is  
> approaching
> > fitting this description.
> >
> > The user-typed code can both create and use static link values  
> that do
> > not occur in the original code that the compiler saws. The simplest
> > example is, when stopped in a nested procedure, the user types
> > 'print NonLocalVariable'. This works fine right now, unless the  
> compiler
> > has not stored the SL in the current proc's activation record,  
> which a
> > compiler might well do if the nested proc, as coded, has no  
> references
> > to any nonlocal variables. (Or keeps the SL only in a register,  
> etc.)
> >
> > You could say this is only a convenience, as the user could work  
> around
> > it by just typing 'up' the exact number of times to get to the  
> static
> > parent's frame (which is, of course, also a dynamic ancestor).  
> But that
> > can slow down debugging a great deal, and I often find it a very  
> important
> > convenience. Anyway, you could say that to a degree about much of  
> what a
> > debugger does.
> >
> > A more complex example might be 'print Nested1 
> (Nested2,NonlocalVar)',
> > where Nested1 and Nested2 are nested procedures. Here, the debugger
> > has create and pass the appropriate SL value in order to call  
> Nested1,
> > which, unless Nested1 is one scope deeper than the current stop  
> point,
> > also entails m3gdb's fetching the SL that was given to some  
> procedure
> > on the current dynamic call chain. The same things are required of
> > m3gdb to construct the closure for Nested2, which it must pass to  
> Nested1.
> > And if NonlocalVar is in an outer scope where Nested1 and/or Nested2
> > aren't visible, there would be no workaround possible at all.
> >
> > Tony, I haven't been able to try out your latest change anyway, as
> > I seem to have once again regressed to not being able to get CM3 to
> > link to a working version of libpthread on Mandriva. Also, I am  
> about
> > to travel for a week, so may be out of this discussion for a while.
> >
> >
> > Tony Hosking wrote:
> > > Rodney,
> > >
> > > I just realised that the static chains variables are *not*  
> always being
> > > set up for nested function internal use. i.e., callers will  
> pass them,
> > > but nested functions that don't explicitly use them will not  
> have a
> > > static chain. Is this something essential for you? Why would  
> you need
> > > a static chain from within a nested function that doesn't use  
> any outer
> > > scope variables? Since it doesn't use any outer scope variables  
> they
> > > cannot affect its execution, so debugging the value of those  
> variables
> > > seems unnecessary. This is the default behavior of gcc for  
> static chains.
> > >
> > > -- Tony
> > >
> > >
> >
> > --
> > -------------------------------------------------------------
> > Rodney M. Bates, retired assistant professor
> > Dept. of Computer Science, Wichita State University
> > Wichita, KS 67260-0083
> > 316-978-3922
> > rodney.bates at wichita.edu
>
>
> Put your friends on the big screen with Windows Vista® + Windows  
> Live™. Start now!