[Fwd: Re: c++/1512: no canonical way to output names of C++ types]

[Fwd: Re: c++/1512: no canonical way to output names of C++ types]

[Andrew Cagney]

[-- Attachment #1: Type: text/plain, Size: 536 bytes --]

>  Fixes for this are in progress on drow-cplus-branch.  All the necessary
>  code is there, it just needs to be hooked up :)  I think I've gotten
>  derailed on some other problems first, though.  The basic problem is
>  that we want to canonicalize the names before adding them to the symbol
>  table, and that increases startup time.  I'm looking for ways to shave
>  startup time elsewhere.  I have a few ideas.

If it is going to be so expensive, can it be put off until later, when 
the canonical value is needed?

enjoy,
Andrew


[-- Attachment #2: Re: c++/1512: no canonical way to output names of C++ types --]
[-- Type: message/rfc822, Size: 3926 bytes --]

From: Daniel Jacobowitz <drow@mvista.com>
To: nobody@sources.redhat.com
Cc: gdb-prs@sources.redhat.com,
Subject: Re: c++/1512: no canonical way to output names of C++ types
Date: 14 Jan 2004 23:18:00 -0000
Message-ID: <20040114231800.3737.qmail@sources.redhat.com>

The following reply was made to PR c++/1512; it has been noted by GNATS.

From: Daniel Jacobowitz <drow@mvista.com>
To: carlton@bactrian.org
Cc: gdb-gnats@sources.redhat.com
Subject: Re: c++/1512: no canonical way to output names of C++ types
Date: Wed, 14 Jan 2004 18:12:13 -0500

 On Wed, Jan 14, 2004 at 06:26:13PM -0000, carlton@bactrian.org wrote:
 > 
 > >Number:         1512
 > >Category:       c++
 > >Synopsis:       no canonical way to output names of C++ types
 > >Confidential:   no
 > >Severity:       serious
 > >Priority:       medium
 > >Responsible:    unassigned
 > >State:          open
 > >Class:          change-request
 > >Submitter-Id:   net
 > >Arrival-Date:   Wed Jan 14 18:28:00 UTC 2004
 > >Closed-Date:
 > >Last-Modified:
 > >Originator:     carlton@bactrian.org
 > >Release:        2004-01-14-cvs
 > >Organization:
 > >Environment:
 > any
 > >Description:
 > We currently don't have any canonical way to output
 > names of C++ types.  E.g. const char * vs.
 > char const *; more subtleties arise when dealing with
 > templates.
 > 
 > This is, of course, less important than the fact that
 > we don't know how to recognize different names
 > of types as equivalent when doing input, but it's
 > still unfortunate.
 
 Fixes for this are in progress on drow-cplus-branch.  All the necessary
 code is there, it just needs to be hooked up :)  I think I've gotten
 derailed on some other problems first, though.  The basic problem is
 that we want to canonicalize the names before adding them to the symbol
 table, and that increases startup time.  I'm looking for ways to shave
 startup time elsewhere.  I have a few ideas.
 
 -- 
 Daniel Jacobowitz
 MontaVista Software                         Debian GNU/Linux Developer

Re: [Fwd: Re: c++/1512: no canonical way to output names of C++ types]

[Daniel Jacobowitz]

On Thu, Jan 15, 2004 at 11:18:22AM -0500, Andrew Cagney wrote:
> > Fixes for this are in progress on drow-cplus-branch.  All the necessary
> > code is there, it just needs to be hooked up :)  I think I've gotten
> > derailed on some other problems first, though.  The basic problem is
> > that we want to canonicalize the names before adding them to the symbol
> > table, and that increases startup time.  I'm looking for ways to shave
> > startup time elsewhere.  I have a few ideas.
> 
> If it is going to be so expensive, can it be put off until later, when 
> the canonical value is needed?

Possibly.  We search for the canonical form in the symbol table, which
means we need to canonicalize before adding them to the (partial, full,
maybe minimal also) symbol table.  The trick will be delaying the
add-to-symbol-table until the symbol is needed, using a clever trick
suggested by Ian - keeping a list of pending symbols by basename,
basically.

There are some complications.  Particuarly, the code to infer classes'
fully qualified names depends on having the canonical demangled name
available.  I may be able to do something about this but I'm not sure
what yet.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: [Fwd: Re: c++/1512: no canonical way to output names of C++ types]

[Andrew Cagney]

> 
> Possibly.  We search for the canonical form in the symbol table, which
> means we need to canonicalize before adding them to the (partial, full,
> maybe minimal also) symbol table.  The trick will be delaying the
> add-to-symbol-table until the symbol is needed, using a clever trick
> suggested by Ian - keeping a list of pending symbols by basename,
> basically.

Is there a pointer to this thread?

Andrew

Re: [Fwd: Re: c++/1512: no canonical way to output names of C++ types]

[Daniel Jacobowitz]

On Thu, Jan 15, 2004 at 11:33:35AM -0500, Andrew Cagney wrote:
> >
> >Possibly.  We search for the canonical form in the symbol table, which
> >means we need to canonicalize before adding them to the (partial, full,
> >maybe minimal also) symbol table.  The trick will be delaying the
> >add-to-symbol-table until the symbol is needed, using a clever trick
> >suggested by Ian - keeping a list of pending symbols by basename,
> >basically.
> 
> Is there a pointer to this thread?

Yes, it was on gcc-patches:
  http://gcc.gnu.org/ml/gcc-patches/2003-12/msg02174.html
and followups.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer