Re: FRAME_ARGS_SKIP

[Andrew Cagney <ac131313 at cygnus dot com>]

> Here's the entry from gdbarch.sh for FRAME_ARGS_SKIP:
> 
> v:2:FRAME_ARGS_SKIP:CORE_ADDR:frame_args_skip::::0:-1
> 
> This sets 0 as the static default, -1 as the predefault, and specifies
> no postdefault. If I'm reading gdbarch.sh correctly, this means that:


gdbarch.log (generated) should confirm this:


> - we use 0 as the value in the dummy gdbarch object we use during
> startup, and


Just FYI, that value was chosen strictly on the basis that it didn't 
appear to make GDB dump core during initialization.


> - we set this field to -1 in a newly allocated gdbarch object, and we
> report an internal error if we notice later that it still has this
> value.
> 
> If I add a postdefault, then instead of getting an internal error when
> it's not initialized, we'll just drop in the postdefault value.
> 
> Is that correct?


Yes.  It is also possible to make the pre-default zero and gag the error 
checking.


> This field should be zero for almost every architecture. Would anyone
> mind if I added a postdefault of zero?


Now that is a US$64 question (rougly AUD$64,000,000).

The strategy, when converting a macro to multi-arch, has been to 
preserve existing behavour.  If, pre-multi-arch, not defining a macro 
caused compiler errors, post- multi-arch it triggered an internal error.

If someone wants to go back and change a variable, after the event, 
then, mostly I'm not worried.  I'd just discourage it when there are 
more complex initialization dependencies such as for the dummy frame.

	Andrew