Re: GDB can't parse variables named "memory" or "array"?

[Paul Smith <paul@mad-scientist.net>]

I filed https://sourceware.org/bugzilla/show_bug.cgi?id=22231


On Fri, 2017-09-29 at 16:17 -0400, Paul Smith wrote:
> On Fri, 2017-09-29 at 14:57 -0400, Sergio Durigan Junior wrote:
> > On Friday, September 29 2017, Paul Smith wrote:
> > > I've tried this with lots of different versions of GDB (7.7.1,
> > > 7.11, 7.12, etc.), and none of them work when debugging my programs
> > > (not just my main program but all my unit tests as well):
> > > 
> > >    (gdb) p memory
> > >    A syntax error in expression, near `'.
> > > 
> > >    (gdb) p array
> > >    A syntax error in expression, near `'.
> > You can also enable "set debug parser on" and/or "set debug
> > expression 1" inside your "faulty" GDB and see if it helps with
> > anything.
> 
> This gave very useful information, actually.  I now see what's
> happening, although I can't understand why no one has noticed this so
> I'm not sure.
> 
> Printing "array" or "memory" shows:
> 
>   (gdb) p memory
>   Starting parse
>   Entering state 0
>   Reading a token: Next token is token FILENAME (bval<0x335a1d0>)
>   Shifting token FILENAME (bval<0x335a1d0>)
>   Entering state 47
>   Reducing stack by rule 107 (line 932):
>      $1 = token FILENAME (bval<0x335a1d0>)
>   -> $$ = nterm block ()
>   Stack now 0
>   Entering state 57
>   Reading a token: Now at end of input.
>   A syntax error in expression, near `'.
> 
> The problem appears to be with the C++ header files, which don't have
> extensions.  Here's a repro case:
> 
> $ cat gdbtest.cpp
> // must include memory
> #include <memory>
> 
> class Foo
> {
>     char* memory;
> };
> 
> Foo foo;
> 
> int main(int, char**)
> {
>     return 1;
> }
> 
> $ g++ --ggdb3 -o gdbtest gdbtest.cpp
> 
> $ gdb -n gdbtest
>  ...
> (gdb) br 13
> (gdb) run
> (gdb) p foo.memory
> A syntax error in expression, near `'.
> 
> 
> Note that you have to use -ggdb3 to see the problem; just using -g
> doesn't show the error.  Also it ends up that the symbol must be part
> of a class (or probably struct but not tested): if it's a global or
> auto symbol it's interpreted correctly.
> 
> Why is GDB even considering a filename to be part of a print
> expression?