[APPROVE] __cxa_pure_virtual()

[APPROVE] __cxa_pure_virtual()

[Bart Veer]

I have just committed a patch to CYGPKG_INFRA to provide an
implementation of __cxa_pure_virtual() that is appropriate for eCos,
overriding the one in libsupc++. 

  http://sources.redhat.com/ml/ecos-patches/2003-03/msg00209.html

Without this patch any C++ code that uses pure virtual functions will
fail to link. The patch adds a single new function
__cxa_pure_virtual() which will never get called in normal operation,
but which does get referenced by the compiled C++ code. The only time
this function may get called is if the application calls a virtual
function while the object is still being created, which gives
undefined behaviour. Hence the risk is small.

Bart

Re: [APPROVE] __cxa_pure_virtual()

[Jonathan Larmour]

Bart Veer wrote:
> I have just committed a patch to CYGPKG_INFRA to provide an
> implementation of __cxa_pure_virtual() that is appropriate for eCos,
> overriding the one in libsupc++. 
> 
>   http://sources.redhat.com/ml/ecos-patches/2003-03/msg00209.html

No objection in principle, although I think you could add a diag_printf 
after the CYG_FAIL saying the same thing so there is some potential 
diagnostic if people don't have asserts enabled.

Perhaps even
#ifdef CYGDBG_USE_ASSERTS
CYG_FAIL
#else
diag_printf
#endif

Whatever the specific implementation, the patch is worth including in 2.0.

Jifl

-- 
eCosCentric    http://www.eCosCentric.com/    The eCos and RedBoot experts
--[ "You can complain because roses have thorns, or you ]--
--[  can rejoice because thorns have roses." -Lincoln   ]-- Opinions==mine

Re: [APPROVE] __cxa_pure_virtual()

[John Dallaway]

I have no objection to this patch for the eCos 2.0 branch.

John Dallaway

-----Original Message-----
From: Jonathan Larmour
Date: Thursday 27 Mar 2003 18:07
Subject: Re: [APPROVE] __cxa_pure_virtual()

> Bart Veer wrote:
> > I have just committed a patch to CYGPKG_INFRA to provide an
> > implementation of __cxa_pure_virtual() that is appropriate for eCos,
> > overriding the one in libsupc++.
> >
> >   http://sources.redhat.com/ml/ecos-patches/2003-03/msg00209.html
>
> No objection in principle, although I think you could add a diag_printf
> after the CYG_FAIL saying the same thing so there is some potential
> diagnostic if people don't have asserts enabled.
>
> Perhaps even
> #ifdef CYGDBG_USE_ASSERTS
> CYG_FAIL
> #else
> diag_printf
> #endif
>
> Whatever the specific implementation, the patch is worth including in
> 2.0.
>
> Jifl

Re: [APPROVE] __cxa_pure_virtual()

[Bart Veer]

>>>>> "Jifl" == Jonathan Larmour <jifl@eCosCentric.com> writes:

    Jifl> Bart Veer wrote:
    >> I have just committed a patch to CYGPKG_INFRA to provide an
    >> implementation of __cxa_pure_virtual() that is appropriate for eCos,
    >> overriding the one in libsupc++. 
    >> 
    >> http://sources.redhat.com/ml/ecos-patches/2003-03/msg00209.html

    Jifl> No objection in principle, although I think you could add a
    Jifl> diag_printf after the CYG_FAIL saying the same thing so
    Jifl> there is some potential diagnostic if people don't have
    Jifl> asserts enabled.

    Jifl> Perhaps even
    Jifl> #ifdef CYGDBG_USE_ASSERTS
    Jifl> CYG_FAIL
    Jifl> #else
    Jifl> diag_printf
    Jifl> #endif

No. __cxa_pure_virtual() should never get called. Even when still
debugging, there are very few applications sufficiently broken that
__cxa_pure_virtual() would get called. In non-debug builds we want a
minimal implementation, and having a diag_printf() in there would be a
waste of code space.

Arguably we should go even further and in non-debug builds alias this
to cyg_assert_fail() or some other dummy function, saving a couple
more bytes.

Bart

-- 
Bart Veer                       eCos Configuration Architect
http://www.ecoscentric.com/     The eCos and RedBoot experts

Re: [APPROVE] __cxa_pure_virtual()

[Jonathan Larmour]

Bart Veer wrote:
>>>>>>"Jifl" == Jonathan Larmour <jifl@eCosCentric.com> writes:
> 
> 
>     Jifl> Bart Veer wrote:
>     >> I have just committed a patch to CYGPKG_INFRA to provide an
>     >> implementation of __cxa_pure_virtual() that is appropriate for eCos,
>     >> overriding the one in libsupc++. 
>     >> 
>     >> http://sources.redhat.com/ml/ecos-patches/2003-03/msg00209.html
> 
>     Jifl> No objection in principle, although I think you could add a
>     Jifl> diag_printf after the CYG_FAIL saying the same thing so
>     Jifl> there is some potential diagnostic if people don't have
>     Jifl> asserts enabled.
> 
>     Jifl> Perhaps even
>     Jifl> #ifdef CYGDBG_USE_ASSERTS
>     Jifl> CYG_FAIL
>     Jifl> #else
>     Jifl> diag_printf
>     Jifl> #endif
> 
> No. __cxa_pure_virtual() should never get called. Even when still
> debugging, there are very few applications sufficiently broken that
> __cxa_pure_virtual() would get called. In non-debug builds we want a
> minimal implementation, and having a diag_printf() in there would be a
> waste of code space.

Fair point.

> Arguably we should go even further and in non-debug builds alias this
> to cyg_assert_fail() or some other dummy function, saving a couple
> more bytes.

Well, in the future we should have a general way to deal with program 
termination/abort whatever. abort() has too many existing semantic 
requirements, but to achieve that sort of effect, particularly when a ROM 
monitor is present and you want to return to it, or whatever other 
(configurable) behaviour. But for much later (beyond 2.1 probably in 
practice).

Jifl
-- 
eCosCentric    http://www.eCosCentric.com/    The eCos and RedBoot experts
--[ "You can complain because roses have thorns, or you ]--
--[  can rejoice because thorns have roses." -Lincoln   ]-- Opinions==mine