Re: [Axiom-developer] Re: root chunks

[Ralf Hemmecke]

> > #line 56 "myalps/prtype.as.nw"

> Yes, that is the best case.  However, I don't know that there is a 
> general way to have the Lisp compiler preserve that information in 
> useful form in the binary.  We can preserve it when we translate to 
> Lisp (or C or whatever - regardless of language) but when that final 
> file is compiled by a compiler we didn't write I'm not sure how to
> make it preserve the information for the debugger.  I'm even less
> sure how to do this portably - we might be able to talk to the GCL
> and SBCL devs, I guess.

I am pretty sure that someone can confirm that gcc has an option that
compiles the #line information into the executable. I am not so sure 
about gcl. Camm?

You don't need to debug with sbcl. If a bug happens via SBCL and not via 
GCL then that is a bug in the compiler and not in the algebra code. So 
you would not be interested in debugging unless you are a compiler writer.

> > There is simply no need to have chunk names corresponding to 
> > filenames.

> At some point I think the Algebra code needs to dictate file names 
> (unless the build somehow records what it needs for loading), in
> order to be able to load the correct binary files corresponding to
> various bits of functionality later.

I don't know what problem you want to solve, but all I can see it that 
there should be a way that the system knows that for domain/category X 
it must load the file Y. To me it looks like a simple translation table 
is sufficient and a user never needs to know about the filename where X 
is defined. We should have a system where you just click on X and your 
editor opens and shows you the right place in file Y. That the compiled 
binary of Y is in Z is only important for the loading mechanism. Why 
would you want to bother the writer of X with specifying a filename for Z?

Ideally the writer of X should not even need to know the name of Y.

Ralf