--enable-__cxa_atexit

--enable-__cxa_atexit

[Scott Robert Ladd]

Since gcc 3.1, I've built the compiler with the --enable-__cxa_atexit option
on Linux/Intel systems, as suggested by the ABI recommendations at:

http://gcc.gnu.org/gcc-3.2/c++-abi.html

I recently installed a Debian "sid" system, which installed a Debian package
for gcc 3.2.2. Checking the option, I discovered that the package was
compiled with --disable-__cxa_atexit.

Can anyone explain why Debian's maintainers made the above choice?

--
Scott Robert Ladd
Coyote Gulch Productions (http://www.coyotegulch.com)
Professional programming for science and engineering;
Interesting and unusual bits of very free code.

Re: --enable-__cxa_atexit

[Daniel Jacobowitz]

On Mon, Feb 10, 2003 at 03:44:11PM -0500, Scott Robert Ladd wrote:
> Since gcc 3.1, I've built the compiler with the --enable-__cxa_atexit option
> on Linux/Intel systems, as suggested by the ABI recommendations at:
> 
> http://gcc.gnu.org/gcc-3.2/c++-abi.html
> 
> I recently installed a Debian "sid" system, which installed a Debian package
> for gcc 3.2.2. Checking the option, I discovered that the package was
> compiled with --disable-__cxa_atexit.
> 
> Can anyone explain why Debian's maintainers made the above choice?

You picked a bad week to check.  It was an error involving pattern
matching and the netbsd-i386 port.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: --enable-__cxa_atexit

[Gabriel Dos Reis]

"Scott Robert Ladd" <scott@coyotegulch.com> writes:

| Since gcc 3.1, I've built the compiler with the --enable-__cxa_atexit option
| on Linux/Intel systems, as suggested by the ABI recommendations at:
| 
| http://gcc.gnu.org/gcc-3.2/c++-abi.html
| 
| I recently installed a Debian "sid" system, which installed a Debian package
| for gcc 3.2.2. Checking the option, I discovered that the package was
| compiled with --disable-__cxa_atexit.
| 
| Can anyone explain why Debian's maintainers made the above choice?


Why don't you ask that on Debian lists?

-- Gaby

RE: --enable-__cxa_atexit

[Scott Robert Ladd]

Gabriel Dos Reis wrote:
> | Can anyone explain why Debian's maintainers made the above choice?
>
> Why don't you ask that on Debian lists?

Because I thought, perhaps, that I misunderstood the need to
include --enable-__cxa_exit in the configuration of gcc. Had I asked in the
Debian lists, I probably would have had someone tell me to ask here... ;)

..Scott

Re: --enable-__cxa_atexit

[Gabriel Dos Reis]

"Scott Robert Ladd" <scott@coyotegulch.com> writes:

| Gabriel Dos Reis wrote:
| > | Can anyone explain why Debian's maintainers made the above choice?
| >
| > Why don't you ask that on Debian lists?
| 
| Because I thought, perhaps, that I misunderstood the need to
| include --enable-__cxa_exit in the configuration of gcc. Had I asked in the
| Debian lists, I probably would have had someone tell me to ask here... ;)

Packagers best know why they include or exclude features.

-- Gaby

RE: --enable-__cxa_atexit

[Scott Robert Ladd]

Gabriel Dos Reis wrote:
> Packagers best know why they include or exclude features.

And the owners of a project probably know best which options are required
for configuration. I wanted to know if --enable-__cxa_exit was still
required for ABI conformance, or if the rules had changed. My question was
about gcc; this is the gcc list.

..Scott

Re: --enable-__cxa_atexit

[Gabriel Dos Reis]

"Scott Robert Ladd" <scott@coyotegulch.com> writes:

| Gabriel Dos Reis wrote:
| > Packagers best know why they include or exclude features.
| 
| And the owners of a project probably know best which options are required
| for configuration. I wanted to know if --enable-__cxa_exit was still
| required for ABI conformance, or if the rules had changed. My question was
| about gcc; this is the gcc list.

You asked why Debian maintainers didn't include a specific feature.
That is more a question about that distributor's  motive than about GCC.
Certainly you're not arguing that each time some has a question about
why his/her vendor/distributor did or didn't something, he asked here...

-- Gaby

RE: --enable-__cxa_atexit

[Scott Robert Ladd]

> You asked why Debian maintainers didn't include a specific feature.
> That is more a question about that distributor's  motive than about GCC.
> Certainly you're not arguing that each time some has a question about
> why his/her vendor/distributor did or didn't something, he asked here...

I've explained my question several times; ONE MORE TIME:

I wanted to know if --enable-__cxa-exit is still required for ABI
conformance.

That is a GCC question. Perhaps I phrased my original question incorrectly
for your taste -- but I've asked the "right" question several times now
(including above), and haven't received a direct answer. I can *infer* that
it is indeed still required, and as such, leave you to have the final word.

I obviously used the -pedantic switch to invoke your aggressive warning
mechanism.

..Scott

Re: --enable-__cxa_atexit

[Neil Booth]

Scott Robert Ladd wrote:-

> I obviously used the -pedantic switch to invoke your aggressive warning
> mechanism.

Be careful it doesn't bite.  It has been known...

Neil.

Re: --enable-__cxa_atexit

[Gerald Pfeifer]

Gaby, Neil,

I hardly ever disagreed with you in the past, but Scott's experience
with this list (at least concerning this issue) really must be quite
frustrating by now.

The question now is ``Is --enable-__cxa_atexit still required for ABI
compliance'', and that really looks like a GCC question to me (which
I'd answer -- if I knew the answer).

Gerald
-- 
Gerald "Jerry"   pfeifer@dbai.tuwien.ac.at   http://www.pfeifer.com/gerald/

Re: --enable-__cxa_atexit

[Neil Booth]

Gerald Pfeifer wrote:-

> Gaby, Neil,
> 
> I hardly ever disagreed with you in the past, but Scott's experience
> with this list (at least concerning this issue) really must be quite
> frustrating by now.
> 
> The question now is ``Is --enable-__cxa_atexit still required for ABI
> compliance'', and that really looks like a GCC question to me (which
> I'd answer -- if I knew the answer).

Sorry Gerald, I have no idea.  I was trying to add a bit of (British?)
(Neil-specific?) humour.

Neil.

Re: --enable-__cxa_atexit

[Gabriel Dos Reis]

Gerald Pfeifer <pfeifer@dbai.tuwien.ac.at> writes:

| The question now is ``Is --enable-__cxa_atexit still required for ABI
| compliance'', and that really looks like a GCC question to me (which
| I'd answer -- if I knew the answer).

Yes, it is still required.

-- Gaby

Re: --enable-__cxa_atexit

[Mark Mitchell]

--On Tuesday, February 11, 2003 12:00:52 AM +0100 Gerald Pfeifer 
<pfeifer@dbai.tuwien.ac.at> wrote:

> The question now is ``Is --enable-__cxa_atexit still required for ABI
> compliance'', and that really looks like a GCC question to me (which
> I'd answer -- if I knew the answer).

It's not required for ABI conformance -- but it is required to get a 
correct C++ compiler, if you're concerned about the exact order of static 
destructors.

-- 
Mark Mitchell                mark@codesourcery.com
CodeSourcery, LLC            http://www.codesourcery.com

RE: --enable-__cxa_atexit

[Scott Robert Ladd]

Mark Mitchell wrote
> It's not required for ABI conformance -- but it is required to get a
> correct C++ compiler, if you're concerned about the exact order of static
> destructors.

Wonderful! I appreciate the answer.

I had, by the way, dug into the compiler source code finally to figure out
exactly what __cxa_atexit actually *did*. I'm curious as to why this is not
the default option, and why it is listed in the documentation as a
requirement for ABI conformance when it is, in fact, related to C++
correctness.

..Scott

RE: --enable-__cxa_atexit

[Mark Mitchell]

--On Tuesday, February 11, 2003 02:11:46 PM -0500 Scott Robert Ladd 
<scott@coyotegulch.com> wrote:

> the default option, and why it is listed in the documentation as a
> requirement for ABI conformance when it is, in fact, related to C++
> correctness.

I'm not sure about the documentation, but I can tell you why it's not
the default.  It's not the default because it uses a C runtime library
interface that was historically not available on lots of systems; until
that interface was universal we didn't want to make it the case that
binaries built with G++ would't work on a system with a slightly older
C library.

It might be that the default should be changed.

-- 
Mark Mitchell                mark@codesourcery.com
CodeSourcery, LLC            http://www.codesourcery.com

Re: --enable-__cxa_atexit

[Paul Jarc]

Mark Mitchell <mark@codesourcery.com> wrote:
> It's not the default because it uses a C runtime library interface
> that was historically not available on lots of systems;

How about letting ./configure test for that interface, and letting
that determine the default?


paul

Re: --enable-__cxa_atexit

[Mark Mitchell]

--On Tuesday, February 11, 2003 02:17:18 PM -0500 Paul Jarc 
<prj@po.cwru.edu> wrote:

> Mark Mitchell <mark@codesourcery.com> wrote:
>> It's not the default because it uses a C runtime library interface
>> that was historically not available on lots of systems;
>
> How about letting ./configure test for that interface, and letting
> that determine the default?

The reason that wasn't done originally is that I could ./configure on
my machine, find the new C library, and then send you binaries that
didn't work on your machine.

I'm *completely* agnostic on this issue, so I really have nothing more
to say.  The distributors and users have a lot more info; we should do
whatever is handiest for them.

-- 
Mark Mitchell                mark@codesourcery.com
CodeSourcery, LLC            http://www.codesourcery.com

Re: --enable-__cxa_atexit

[Gabriel Dos Reis]

Mark Mitchell <mark@codesourcery.com> writes:

| --On Tuesday, February 11, 2003 12:00:52 AM +0100 Gerald Pfeifer
| <pfeifer@dbai.tuwien.ac.at> wrote:
| 
| > The question now is ``Is --enable-__cxa_atexit still required for ABI
| > compliance'', and that really looks like a GCC question to me (which
| > I'd answer -- if I knew the answer).
| 
| It's not required for ABI conformance -- but it is required to get a
| correct C++ compiler, if you're concerned about the exact order of
| static destructors.

Then, the documentaion needs to be corrected.

-- Gaby