From mboxrd@z Thu Jan  1 00:00:00 1970
From: Michael Elizabeth Chastain <chastain@cygnus.com>
To: chastain@cygnus.com, msnyder@cygnus.com
Cc: gdb-patches@sources.redhat.com
Subject: Re: [RFA] testsuite/gdb.c++/cplusfuncs.{exp,cc}: work with either g++ demangler
Date: Mon, 12 Feb 2001 12:30:00 -0000
Message-id: <200102122030.MAA22848@bosch.cygnus.com>
X-SW-Source: 2001-02/msg00176.html

Hi Michael,

> I'm curious why you didn't simply write separate regular expressions,
> one to recognize the new demangler style and one to recognize the old,
> rather than add all this new mechanics.

That's a reasonable question.

The script does not actually know about about "v2 demangler" and "v3
demangler" and which regexp's go with which version.  I wrote it more
like autoconf (although I am not experienced with autoconf).  It probes
for each type separately and accommodates all the versions that I
have ever seen.  If tomorrow's demangler wants to print 'foo    &',
or report every 'int' as a 'signed int', it would be easy to maintain
the script.

> I'm concerned about the tcl scripts becoming complicated and difficult
> to maintain.

There are two parts.  The probing part is meant to be independent and
can be copied-and-pasted into other scripts or put into a library.

The probing part can also be reduced to nothing if we stopped supporting
v2 abi.  The strings are initialized to their v3 values.

As far as the use goes -- it does bother me to fill up the whole script
with "$dm_type_foo_ref".

Part of the problem is that a lot of these test scripts are not very
careful about "input to gdb" (which must always be a literal string)
versus "output from gdb" (which they can model with either a literal
string or with a regular expression).  cplusfuncs.exp likes to specify
one string for both input and output, so it *has* to be a literal.

Michael