Re: Frysk presentation

[Petr Machata <pmachata at redhat dot com>]

Sami Wagiaalla wrote:
> More specifically behdad was excited about ftrace and has made some
> specific requests:
> - To be able to easily specify which library he wants traced
Tracked here: http://sourceware.org/bugzilla/show_bug.cgi?id=5850

> - To be able to trace private functions.
What does this mean?  Is it like he wants to request tracing of private
methods of some class?  E.g. something like:
  -sym='\private(someclass::*@libsoname.*)'
(The \private qualifier doesn't work, I'm just playing around with some
syntax.)

Or if private function is just function not exported to dynamic symbol
table, that would be -sym='@libsoname.*', but then symbol table is
necessary.  This limitation is going to disappear once ftrace is
rewritten to use frysk own framework for breakpoint handling, instead of
reimplementing half of libdwfl on its own :)

> - To be able to tell frysk how to print cairo types.

... and for that matter, ftrace needs smarts to know what prototype a
function has, so that it can format the output accordingly, à la strace.
 Most of that can be derived from debuginfo, but not everything.
Sometimes 'unsigned int' is actually set of or-ed flags, or it accepts
some #defined constants, some pointers are actually return values, etc.
 This all should be possible to configure.

Ideally these configuration files would be distributed in devel
subpackage of the library whose interfaces it describes.

Also ideally the smarts that infer the prototype would be shared with
whoever needs them (dltrace and ltrace come to mind).

PM

Attachment: signature.asc
Description: OpenPGP digital signature