[cmucl-imp] fdefns for flets

[Raymond Toy]

On 1/21/11 1:33 PM, Helmut Eller wrote:
> * Raymond Toy [2011-01-21 15:42] writes:
>
>>> I think this is only generated by genesis.  Or is there another way how
>>> such names can end up in globaldb?
>> I'm sorry, I don't quite follow.  What is only generated by genesis?
> The fdefn objects for flet/labels.  fdefn objects are needed for those
> functions that are called by name.  As symbols in CMUCL have no function
> slot the place to store named functions are fdefn objects.
>
> The loader needs to find the fdefn object for a symbol so somebody has
> to put it in globaldb first.  Usually that's done by %%defun but %%defun
> is not a options during genesis so the compiler additionally emits
> fop-fset.  fop-fset is a nop after genesis.
>
> fop-fset is emitted in c::dump-one-entry and it's guarded by 
>
>   (when (and name (or (symbolp name) (listp name)))
>     ...
>
> flet/labels don't need fdefns because those are always called with
> local-call.  Those could probably be filtered out.  For trace, we
> probably only need the global name; with that we can find the code
> component and then scan the functions in the component for a matching
> flet/labels name.
I changed the above code to skip over list names starting with
flet/labels.  No problem with that.  But now tracing doesn't work.  So
trace will have to grovel the code component to find the name.   That
would take some work to do; I'm not that familiar with the code
component, and it's been a while since I've looked at that.

Ray