Re: RFA: More concise errors for missing shared libraries

[Mark Kettenis <mark.kettenis@xs4all.nl>]

> Date: Tue, 9 May 2006 13:44:26 -0400
> From: Daniel Jacobowitz <drow@false.org>
> 
> One of the quirks of SymbianOS compared to a traditional hosted target is
> that it's quite likely that GDB will have no symbol file at all for many of
> the loaded libraries.  A ROM image may contain many DLLs, presented as a
> virtual filesystem, which are effectively preloaded into the memory space of
> every process.  A user probably only has symbol files for his own DLLs, not
> the DLLs which provide operating system services.  So, the behavior of GDB
> when shared libraries can not be found is much more important than on other
> targets.

Did you consider providing stub ELF libraries for the system DLL's too?

> Right now this is really noisy.  We get a two-line error message for each
> library from update_solib_list when we detect the new library, and another
> one from solib_read_symbols later.  This patch condenses those into a single
> warning:
> 
> warning: Could not load shared library symbols for 2 libraries, e.g. /lib/libnss_dns.so.2.
> Do you need "set solib-search-path" or "set solib-absolute-prefix"?

I think this is bad.  I really hate it when tools give me incomplete
error information.  Consider the situation where I have set
solib-search-path/solib-absolute-prefix, but only have a subset of the
libraries available.  I go and install libnss_dns.so.2, but I have no
clue what the other library is.  I have to rerun the program in GDB to
find out the name of the other library.

To reduce the clutter, I think we should not print the

   Do you need "set solib-search-path" or "set solib-absolute-prefix"?

suggestion for every shared library, but only once, preferably at the
end of of all output.

If that would still provide too much clutter, we should perhaps
introduce a user-settable limit for these.

Mark