[Crash-utility] [PATCH] displaying anonymous members of struct/union

[Dave Anderson]

----- Original Message -----
> At 2012-3-6 0:25, Dave Anderson wrote:
> >
> >
> > ----- Original Message -----
> >> Hello Dave,
> >>
> >> I have realized the function to display anonymous members of
> >> struct/union. There are three changes, which are listed below, and I
> >> will take "struct page" for example.
> >>
> >> 1. anonymous members will be showed when using "struct page"
> >
> > That's nice...
> >
> >> 2. anonymous member, "struct page.private", will be showed together with
> >> its offset
> >
> > Sorry -- but doing that redefines the "struct" command output for both named
> > members and anonymous members.  The "struct -o" option should be used to prepend
> > the offset.  Can you re-work the patch so that it behaves the same way
> > as it currently does?
> >
> >> 3. anonymous member with the address of the structure, "struct page.private xxx", will be showed.
> >>
> >> About the implementation, gdb code is changed, which is used to change
> >> the output of the gdb command, "ptype ...". And function
> >> "arg_to_datatype" is also modified to analyze anonymous member.
> >
> > I'm presuming (but not sure) that the change to anon_member_offset() to use
> > open_tmpfile2() is because your patch introduces a new potential path to
> > anon_member_offset() that would be inside a open_tmpfile()/close_tmpfile()
> > region?  Is that correct?
> 
> Exactly, when showing struct.member, function do_datatype_declaration()
> is involved. And in this function, open_tmpfile() is called. Then after
> the call of open_tmpfile() and before the call of close_tmpfile(),
> show_member_offset(), in which I added ANON_MEMBER_OFFSET, is called.
> This macro will call anon_member_offset(). So I change to use
> open_tmpfile2() to prohibit to recursive usage.
> 
> >
> > The change to gdb's c_type_print_base() function is interesting -- do
> > you consider that a bug in gdb, or do you think they have a reason
> > for doing it the way it's done now?  I ask because gdb has several
> > structure-printing optional variables, so I'm kind of surprised that
> > it's hard-wired to behave the way it does.
> 
> The original gdb will display at most three levels of structure in depth
> when using command "ptype ..." without concerning the anonymous member.
> In my patch, I modified the calculation of "show", which is used to
> indicate the level. The show will not decrease 1, using my patch, when
> anonymous structure appears.
> 
> Take the below struct for example:
> 
> type = struct a {
>      struct {
>          struct {
>              union {
>                  struct {
>                  ...
>                  };
>                  ...
>              };
>          };
>      };
> };
> 
> All will be displayed using my patch. And the original gdb will only
> display like:
> 
> type = struct a {
>      struct {
>          struct {...}
>      };
> };
> 
> The original gdb, in my opinion, shows simplicity. And comes to the
> aspect of source code, nothing obviously shows the differentiation.
> These are all that I can figure out until now.

Yeah, I guess that's what they had in mind, although preferring simplicity 
over completeness just seems a little strange for a debugger... 

In any case, thanks for digging into this.  Your patch is very useful, 
and it is queued for crash-6.0.5.

Thanks,
  Dave