[Ltrace-devel] missing library traces

[Petr Machata]

"Oliver Metz" <oliver at freetz.org> writes:

> Is this a ltrace bug? What information can I provide to help to fix
> it? Or does our toolchain misbehave? I read some mails from Sedat in
> the archives about the PLT topic but nothing the helped me to solve
> this issue. You can find some more information in our Trac ticket
> (http://freetz.org/ticket/2326).

It might be a missing feature.  ltrace generally relies on existence of
PLT slots, through which all call to dynamic libraries are made.  We put
breakpoints to these slots, so as to catch these calls.  If there are no
PLT slots, we can't trace library calls.

In practice, on PowerPC in particular, PLT slots themselves are not
really useful for these breakpoints and considerable hackery is
necessary to achieve the effect "as if" we were tracing PLT calls.
Perahaps MIPS backend could do something unusual to trace the -mno-plt
case as well.  It all depends on how tricky it would be to support, and
whether it's the common case.  From a couple Google hits, I gather -mplt
really is the usual case, but maybe that's different for embedded
systems.

As they say, "we accept patches" ;)

Thanks,
PM