Re: [just for the record]: new prologue analyzer for S/390

[Daniel Jacobowitz <drow at mvista dot com>]

On Mon, Apr 21, 2003 at 12:31:34PM -0400, Andrew Cagney wrote:
> >I think this patch shouldn't be committed; I'm just posting it for
> >reference.
> >
> >This patch implements a new prologue analyzer for the S/390.  It's
> >meant to be general enough to handle the complex prologues GCC emits
> >on the S/390, and robust enough to tolerate compiler changes.  In my
> >experience, it does pretty well, even on optimized code.
> >
> >However, the S/390 GDB folks at IBM and I agree that GDB on the S/390
> >should move towards using Dwarf 2 CFI and location lists whenever
> >possible, and do only minimal prologue analysis to handle those few
> >common cases where Dwarf 2 CFI is not available.  And it looks to me
> >as if the work needed in GDB's core code to make it possible for any
> >target to use Dwarf 2 CFI is almost complete.  In that light, it
> >doesn't make sense to dump a new, large, complex prologue analyzer
> >into the code base that will soon be eclipsed by a better solution.


> - the need to have gdb behave reasonably well when there is no, or 
> minimal, debug info

The others aside for the moment, I believe this is the best reason to
continue to support the prologue analyzers.

On the other hand, we're making active progress in providing more and
better CFI; and now we support debug info in a separate file from the
code; so some day soon I expect that I won't even care how GDB behaves
without debug info except for debugging other people's proprietary
software to which I don't have source.  And in that case (which comes
up occasionally) the level of sophistication needed from a prologue
analyzer is extremely low.


-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer