objc-lang.c portability (Was Re: [PATCH] Step over Objective-C dispatch function)

objc-lang.c portability (Was Re: [PATCH] Step over Objective-C dispatch function)

[Adam Fedor]

On Thursday, March 20, 2003, at 02:08 PM, Andrew Cagney wrote:
>
> I've looked at what the underlying code is trying to do and, 
> unfortunatly, the original objc-lang.c botched its portability(1), 
> sigh!  The file is currently native only so infrun.c can't directly 
> refer to objc-lang.c, and hence, will need to go via a dispatch table. 
>  Going via a dispatch table wouldn't hurt anyway.
>
> (1) I noticed that the parameter extract methods assume host=target.
>

I'm wondering what's not portable with objc-lang.c? With the remaining 
patches I have that require this, I was thinking it might be easier to 
just fix it now. The alternative appears to be adding a whole bunch of 
things to the language_defn structure. I admit I don't really 
understand the problem though.

Re: objc-lang.c portability (Was Re: [PATCH] Step over Objective-C dispatch function)

[Daniel Jacobowitz]

On Wed, Mar 26, 2003 at 04:31:23PM -0700, Adam Fedor wrote:
> 
> On Thursday, March 20, 2003, at 02:08 PM, Andrew Cagney wrote:
> >
> >I've looked at what the underlying code is trying to do and, 
> >unfortunatly, the original objc-lang.c botched its portability(1), 
> >sigh!  The file is currently native only so infrun.c can't directly 
> >refer to objc-lang.c, and hence, will need to go via a dispatch table. 
> > Going via a dispatch table wouldn't hurt anyway.
> >
> >(1) I noticed that the parameter extract methods assume host=target.
> >
> 
> I'm wondering what's not portable with objc-lang.c? With the remaining 
> patches I have that require this, I was thinking it might be easier to 
> just fix it now. The alternative appears to be adding a whole bunch of 
> things to the language_defn structure. I admit I don't really 
> understand the problem though.

At the very least FETCH_ARGUMENT and CONVERT_FUNCPTR need to go into
the architecture vector.  They're completely incorrect:
  - they assume host == target
  - they will prevent GDB from compiling on host !< {i386, ppc, sparc, hppa}

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: objc-lang.c portability (Was Re: [PATCH] Step over Objective-C dispatch function)

[Andrew Cagney]

> 
> On Thursday, March 20, 2003, at 02:08 PM, Andrew Cagney wrote:
> 
> I've looked at what the underlying code is trying to do and, unfortunatly, the original objc-lang.c botched its portability(1), sigh!  The file is currently native only so infrun.c can't directly refer to objc-lang.c, and hence, will need to go via a dispatch table.  Going via a dispatch table wouldn't hurt anyway.
> 
> (1) I noticed that the parameter extract methods assume host=target.
> 
> 
> I'm wondering what's not portable with objc-lang.c? With the remaining patches I have that require this, I was thinking it might be easier to just fix it now. The alternative appears to be adding a whole bunch of things to the language_defn structure. I admit I don't really understand the problem though.

Both getting objc-lang.c portable and extending the language vector are 
things that will, one day, need to be done :-/  My guess is that it will 
be quicker, and lower risk, to attack the language vector side of the 
problem then to more slowly audit / fix the portability problems.  This 
will at least make it possible to enable objc for one or two key native 
systems (e.g., i386 GNU/Linux).

Language vector additions are largly a case of changing this:

	switch (current_language)
	  case C: do something C like;
	  case C++: do something C++ like;
	  ...

into something like:

	switch (current_language)
	  ...
	  default:
  	    current_language->do something language like;

which is very mechanical.  There is plenty of prior art so approval is 
pretty mechanical.  Contrast it to the problem DanielJ pointed out (and 
I hinted at):

> At the very least FETCH_ARGUMENT and CONVERT_FUNCPTR need to go into
> the architecture vector.

As they stand, these new methods won't go into the architecture vector. 
  Adding a new architecture method requires care and consideration, and 
for the above, they will definitly need change.

Andrew