[PATCH] Demangle function names when disassembling

[Tom Tromey]

Andrew> It turns out that the bug that originally caused GDB to display
Andrew> demangled names only applied for some languages, Ada, didn't have
Andrew> this issue, and so has always displayed the mangled names.

Ada is special here ... but I am not sure there's a very good reason for
it.  I've been meaning for a while to see if this difference can be
removed.  Mostly, I think this amounts to demangling Ada names during
symbol reading.

Andrew>     So my initial thought was, lets make 'set print asm-demangle' be on by
Andrew>     default.  Then we've still fixed the bug in GDB (about storing the
Andrew>     mangled vs demangled symbols), but the behaviour of GDB doesn't
Andrew>     actually change (vs 9.x release).

It seems like we could still do this.

Andrew>     First, and maybe the smallest problem, boolean variables have the
Andrew>     "feature" that a user can just type 'set print asm-demangle', which is
Andrew>     equivalent to 'set print asm-demangle on'.  However, this doesn't work
Andrew>     for auto-booleans, in that case you must type 'set print asm-demangle
Andrew>     on'.
    
I think we could just safely ignore this.  I wasn't even aware of this
feature; and anyway it's easy to work around; and finally nobody has
added a asm-demangle setting in 10 years or so due to the bug.

Andrew>     If I flip the default for 'asm-demangle' on the grounds that I don't
Andrew>     think we should suddenly change GDB's behaviour then this would only
Andrew>     be for C++, for Ada I would be suddenly changing the default.
    
I think this is probably fine as well.  I think it makes sense for the
option to do what the documentation says it will do.  Users, if they
notice this, can change the setting.

Yet another option would be to change my patch up-thread to use
"demangle" for the header rather than "asm-demangle".  I tend to think
that would be more confusing, but I figured I'd mention it.

I looked into the history of "asm-demangle", but it seems to predate the
sourceware import.

Andrew>     [...] in some places 'asm_demangle' is completely
Andrew>     independent, and in other places 'asm_demangle' actually overrides
Andrew>     'demangle' rather than deferring to it.
    
Bleah.  Can / should we fix this.

There's also "set demangle-style", which is both obsolete and inherently
broken.  It should be removed.

Andrew> +  if (asm_demangle == AUTO_BOOLEAN_AUTO)
Andrew> +    fprintf_filtered (file,
Andrew> +		      _("Demangling of C++/ObjC names in disassembly "

We should probably use this opportunity to drop the C++/ObjC text, since
this applies to any language with mangling.

Tom