Re: PSA: Formalizing the glibc<->gdb probe-based interface.

Re: PSA: Formalizing the glibc<->gdb probe-based interface.

[Tom Tromey]

>>>>> "Carlos" == Carlos O'Donell <carlos@redhat.com> writes:

Carlos> We document *other* probes in manual/probes.texi, but we should
Carlos> add a section for stable guaranteed probes, and discuss a way
Carlos> to deprecate the probes should we ever need to do that.

Because probes are introspectable, it's actually better to remove them
than to change their semantics -- the tools can adapt more easily to
this.  Of course it's good to have a transition period, or at least some
sort of warning.

As I recall we had some thoughts about ABI compatibility when adding the
probes; namely that it is ok to add arguments to the end, but not ok to
remove arguments or to change the meaning of existing arguments.

HTH,
Tom

Re: PSA: Formalizing the glibc<->gdb probe-based interface.

[Carlos O'Donell]

On 2/13/19 12:13 PM, Tom Tromey wrote:
>>>>>> "Carlos" == Carlos O'Donell <carlos@redhat.com> writes:
> 
> Carlos> We document *other* probes in manual/probes.texi, but we should
> Carlos> add a section for stable guaranteed probes, and discuss a way
> Carlos> to deprecate the probes should we ever need to do that.
> 
> Because probes are introspectable, it's actually better to remove them
> than to change their semantics -- the tools can adapt more easily to
> this.  Of course it's good to have a transition period, or at least some
> sort of warning.
> 
> As I recall we had some thoughts about ABI compatibility when adding the
> probes; namely that it is ok to add arguments to the end, but not ok to
> remove arguments or to change the meaning of existing arguments.

Agreed.

In general I think the probe name should always change if:
* semantics change.
* # of arguments change.

Then you'll always be safe.

If you can't keep implementing the probe it is safest to remove it.

Talk to the users of the probe first and reach consensus :-)

-- 
Cheers,
Carlos.