Re: [PATCH 3/4] gdb: add new -group-by-binary flag to info sources command

[Andrew Burgess <andrew.burgess@embecosm.com>]

* Simon Marchi <simon.marchi@polymtl.ca> [2021-05-13 11:05:44 -0400]:

> On 2021-04-26 1:07 p.m., Andrew Burgess wrote:
> > Currently the 'info sources' command lists all of the known source
> > files together, regardless of their source, e.g. here is a session
> > debugging a test application that makes use of a shared library:
> > 
> >   (gdb) info sources
> >   Source files for which symbols have been read in:
> > 
> >   /tmp/info-sources/test.c, /usr/include/stdc-predef.h,
> >   /tmp/info-sources/header.h, /tmp/info-sources/helper.c
> > 
> >   Source files for which symbols will be read in on demand:
> > 
> >   (gdb)
> > 
> > In this commit I add a new flag to the 'info sources' command,
> > '-group-by-binary'.  When this flag is provided the results are
> > grouped by the binary file that uses that source file.  Here's the
> > same session using the new flag:
> > 
> >   (gdb) info sources -group-by-binary
> >   /tmp/info-sources/test.x:
> > 
> >   /tmp/info-sources/test.c, /usr/include/stdc-predef.h,
> >   /tmp/info-sources/header.h
> > 
> >   /lib64/ld-linux-x86-64.so.2:
> >   (Full debug information has not yet been read for this file.)
> > 
> >   system-supplied DSO at 0x7ffff7fcf000:
> >   (Full debug information has not yet been read for this file.)
> > 
> >   /tmp/info-sources/libhelper.so:
> > 
> >   /tmp/info-sources/helper.c, /usr/include/stdc-predef.h,
> >   /tmp/info-sources/header.h
> > 
> >   /lib64/libc.so.6:
> >   (Full debug information has not yet been read for this file.)
> > 
> >   (gdb)
> > 
> > Notice that in the new output some source files are repeated,
> > e.g. /tmp/info-sources/header.h, as multiple binaries use this source
> > file.
> > 
> > All of the existing regular expression based filtering that exists for
> > 'info sources' still works with the new option.
> 
> In my opinion the new output is just better, more structured.  Do we
> even need to hide it behind an option?  Can we just... change it and
> that's it?  I'd lke if we didn't have to keep the extra complexity
> unnecessarily.

I'm not against changing the output, but I thought as a rule we didn't
significantly change existing command behaviour unless we could show
that the existing behaviour was wrong.

I guess how do folk feel about changing this behaviour?

Thanks,
Andrew