On Tue, 2009-12-01 at 11:27 +0100, Michael Schwingen wrote:
> Zach Welch wrote:
> >> Hm - I'm with David here: I am not very fond of re-inventing parts of 
> >> gdb to include it in OpenOCD.
> >>
> >> Fully implementing this would make OpenOCD depend on libbfd just for 
> >> crash reports - this is ridiculous.
> >>     
> >
> > You clearly missed the part where I say it will be selectable, and
> > "none" would also be an option.  *poof*  It could disappear.
> >   
> But then "none" would have to be the default for distributions - which 
> means you would gain nothing for user-submitted crash reports.

Well, none could still mean glibc, which is all that is used now.
I do not see a reason to strip out any glibc-based functionality on
systems using it, unless a particular platform lacks those features.
IIRC, the actual checks right now would be for dlfcn.h and execinfo.h.
Eventually, the 'none' option would be for only those platforms that
lack an equivalent native solution that requires no new dependencies.

To be perfectly honestly, the actual reason we might gain nothing is
that most distributions strip their binaries of debugging symbols, so
not even GDB would be able to do much with them in the default case.
There are usually ways around that, including rebuilding OpenOCD
unstripped using the native package manager (e.g. Gentoo) or installing
a separate package that contains the debugging information (e.g.
OpenEmbedded, Debian).  

To be fair, these extra steps also moot my Heisenbug argument; however,
these are still activities that could be expected by these platforms'
users.  Running GDB is not a user activity, except _possibly_ when using
it _with_ OpenOCD.  Remember, not everyone uses OpenOCD with GDB.

I consider our "users" to include those developers who want to use
OpenOCD as a replacement for an off-the-shelf tool.  Such users do not
want to be told to debug OpenOCD with GDB, but they might be convinced
to install a debugging version of it (e.g. from Git).  Too many users
might give up when they hear that they need to run GDB to debug the
feature that causes a crash for them.  That goes for any package where
They Just Wanted To Use It.

As challenging as this work can be at times, they may give up for other
reasons later anyway... why make bug reporting any harder for users than
it absolutely needs to be?

Cheers,

Zach
_______________________________________________
Openocd-development mailing list
Openocd-development@lists.berlios.de
https://lists.berlios.de/mailman/listinfo/openocd-development

Reply via email to