Re: [commit] Improved linker-debugger interface

[Sergio Durigan Junior <sergiodj at redhat dot com>]

On Wednesday, June 26 2013, Tom Tromey wrote:

> It seems to me that we could make the warning more verbose and have it
> request that users file a bug report; and it could include a little
> explanation, plus some text to report.  Like:
>
>     warning: Probes-based dynamic linker interface failed:
>       Unknown numeric token ...
>     This means there is a bug, either in gdb or elsewhere in the
>     toolchain.  Please report it to the gdb bugzilla, along with
>     this information:
>       Arch: Whatever
>       Probe name: ...
>       Argument number: ...
>       Argument text: ...
>
> Being extra wordy is a bit of a pain, maybe, since presumably users will
> see it often.  OTOH, it's a "shouldn't happen" scenario where gdb has
> lots of information already...

Sounds good to me, but...

> Also, I wonder why we're trying to use the probes on a platform to which
> the gdb side hasn't been ported.  Are we just optimistically trying to
> parse the assembly operands here?

That's a relevant question.  GDB has a generic asm parser (implemented
in stap-probe.c), which makes use of specific target-dep
functions/variables to correctly handle the idiosyncrasies of each type
of asm.

So far, we have support for x86, x86_64, PPC, PPC64, ARM, and a pending
patch for IA-64.  However, there must be more architectures where we
will need to implement it.  The pre-requisite I am/was using is: if
systemtap-sdt-devel is installable on the system, then GDB needs to
support SDT probes.  Unfortunately I failed to check IA-64, but that is
hopefully fixed now.

OTOH, if the user successfuly installs <sys/sdt.h> on her system, then
GDB should at least be prepared to avoid trying to parse something it
doesn't fully understand.

What could be done is to check gdbarch_stap_is_single_operand_p.  This
is a mandatory function that needs to be set by the target.  If it
doesn't exist, then we know that the target lacks SDT support.

Before I work on a patch, I'd like to know what you think.

Thanks,

-- 
Sergio