HOWTO move a target from old style multiarched to new style multiarched?

HOWTO move a target from old style multiarched to new style multiarched?

[Corinna Vinschen]

Hi,

there were a lot of changes in the last couple of months to convert gdb
to a new way of doing things.  That's probably good but I don't know how
many people actually understand the whole process.  I must admit, I don't.

I'm under the impression the documentation is missing an important part.
There's some sort of description spread over the source code now but it
doesn't give a coherent picture and I also didn't find that in the docs.

Wouldn't it be possible to write a HOWTO for the poor developer soul,
who's left behind with a target with now using 80% deprecated functions?
IMHO, that HOWTO written by somebody who knows what's going on could
tells us

  The datastructure `struct bits' is replaced by `struct pieces'.
  This requires to use a new set of functions...

  Create functions foo(), bar() and baz(), doing...

  If you use function new_and_shiny() you must also replace
  old_and_dirty(), otherwise...

  The functions old() and cruft() are replaced by glitter() which
  returns yadda...

  Function ugly() is not used anymore and can be removed without
  substitution...

and so on.  That would be very helpful and probably speed up the
conversion of targets to "new style" enormously.

Or did I just miss the documentation on this?

Corinna

-- 
Corinna Vinschen
Cygwin Developer
Red Hat, Inc.
mailto:vinschen@redhat.com

Re: HOWTO move a target from old style multiarched to new style multiarched?

[Eli Zaretskii]

> Date: Wed, 11 Jun 2003 14:02:12 +0200
> From: Corinna Vinschen <vinschen@redhat.com>
> 
> there were a lot of changes in the last couple of months to convert gdb
> to a new way of doing things.  That's probably good but I don't know how
> many people actually understand the whole process.  I must admit, I don't.
> 
> I'm under the impression the documentation is missing an important part.
> There's some sort of description spread over the source code now but it
> doesn't give a coherent picture and I also didn't find that in the docs.
> 
> Wouldn't it be possible to write a HOWTO for the poor developer soul,
> who's left behind with a target with now using 80% deprecated functions?

Full agreement.  I think such documentation should be in
gdbint.texinfo.

Re: HOWTO move a target from old style multiarched to new style multiarched?

[Andrew Cagney]

> Hi,
> 
> there were a lot of changes in the last couple of months to convert gdb
> to a new way of doing things.  That's probably good but I don't know how
> many people actually understand the whole process.  I must admit, I don't.

This was my most recent guess:
http://sources.redhat.com/ml/gdb/2003-05/msg00399.html
Briefly, slow eliminate each set_gdbarch_deprecated* call, the rest will 
follow.

> I'm under the impression the documentation is missing an important part.
> There's some sort of description spread over the source code now but it
> doesn't give a coherent picture and I also didn't find that in the docs.

There isn't a single picture as there isn't a single change.  Instead 
there are three largely independant changes:

- frames (the obvious one)
- inferior function calls
- global buffer -> regcache or frame

> Wouldn't it be possible to write a HOWTO for the poor developer soul,
> who's left behind with a target with now using 80% deprecated functions?
> IMHO, that HOWTO written by somebody who knows what's going on could
> tells us

To speak personally, I think it is dangerous for someone to post a HOWTO 
on something they haven't done :-/  I, for instance, have only done the 
d10v and there my only recommendation is to not follow my lead.  I have 
posted several theoretical guidelines but there, it's strictly 
conjecture, and the suggestions haven't received any feedback.

Contrast this to multi-arch where the process was documented after I'd 
done several [partial] multi-arch conversions -> I knew it worked.

Anyway, here is an expansion of the last post:

for method in set_gdbarch_deprecated* do
   look in gdbarch.sh to see if there is a direct replacement
   ex: UNWIND_SP is a direct replacement for TARGET_READ_SP
   look in the d10v for an example
(this should be pretty obvious, replace A with B)

for method in set_gdbarch_deprecated* do
   look in gdbarch.sh to see if it is a dummy-frame related function
   ex: The dummy call frame SP should be set by push_dummy_call
   ex: This is simply not needed. ...
   move, implement, or delete the relevant code

cut a USER_CPU_frame branch
- implement a traditional unwinder so that the basics are working
- implement additional unwinders for sigtramps and the like
- replace the lot

But remember, I don't know if this process works, I've not tried it.

As for doco, the bit I acknowledge is missing is an architectural 
overview of the frame code.  The regcache does have posted doco:
http://sources.redhat.com/ml/gdb/2002-07/msg00202.html

Andrew


>   The datastructure `struct bits' is replaced by `struct pieces'.
>   This requires to use a new set of functions...
> 
>   Create functions foo(), bar() and baz(), doing...
> 
>   If you use function new_and_shiny() you must also replace
>   old_and_dirty(), otherwise...
> 
>   The functions old() and cruft() are replaced by glitter() which
>   returns yadda...
> 
>   Function ugly() is not used anymore and can be removed without
>   substitution...
> 
> and so on.  That would be very helpful and probably speed up the
> conversion of targets to "new style" enormously.
> 
> Or did I just miss the documentation on this?
> 
> Corinna
> 
> -- Corinna Vinschen Cygwin Developer Red Hat, Inc. mailto:vinschen@redhat.com