You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@tuscany.apache.org by Geoffrey Winn <ge...@googlemail.com> on 2006/08/30 10:40:25 UTC

[SDO for C++] Opinions Requested: Adopting stdcxx as C++ library

There is an Apache incubator project to develop an implementation of the
standard C++ library (their web site is here
http://incubator.apache.org/stdcxx<http://incubator.apache.org/stdcxx/#platforms>).
Currently, in SDO for C++, we use the library provided by whichever platform
we happen to run on. Using the Apache stdcxx library instead would provide
us with a number of benefits

- It is available on a wide range of platforms
- It is the same implementation on all of them (so avoids obscure bugs from
slight differences in C++ libraries)
- It is open source so we can contribute fixes rather than writing around
problems.
- We become better citizens of the Apache community

The one difficulty is that once SDO links against the stdcxx library then
all users of SDO must also do so. I think this gives us two options

1. Just do it, and live with the consequences. In this case we will (one way
or another) pre-req stdcxx on all platforms, and all users of SDO for C++
will be required to use stdcxx as their C++ standard library.

2. Create a build time switch that chooses between whatever the platform
offers (ie the current arrangement) and stdcxx. Presumably defaulting to the
current arrangement.

I prefer option 2 but obviously it somewhat complicates our build process
and perhaps more seriously adds another complication to our test cases.

What does the team think?

Geoff.

Re: [SDO for C++] Opinions Requested: Adopting stdcxx as C++ library

Posted by Geoffrey Winn <ge...@googlemail.com>.
OK. That's pretty conclusive. I'll raise a JIRA for it and implement option
2.

Geoff.

On 30/08/06, Jean-Sebastien Delfino <js...@apache.org> wrote:
>
> Simon Laws wrote:
> > On 8/30/06, Pete Robbins <ro...@googlemail.com> wrote:
> >>
> >> On 30/08/06, Andrew Borley <aj...@gmail.com> wrote:
> >> >
> >> > On 8/30/06, Geoffrey Winn <ge...@googlemail.com> wrote:
> >> > <snip>. Using the Apache stdcxx library instead would provide
> >> >
> >> > > us with a number of benefits
> >> >
> >> >
> >> > Agreed. +1 for this.
> >>
> >>
> >> yup!
> >>
> >> The one difficulty is that once SDO links against the stdcxx library
> >> then
> >> > > all users of SDO must also do so. I think this gives us two options
> >> > >
> >> > > 1. Just do it, and live with the consequences. In this case we will
> >> (one
> >> > > way
> >> > > or another) pre-req stdcxx on all platforms, and all users of SDO
> >> for
> >> > C++
> >> > > will be required to use stdcxx as their C++ standard library.
> >> > >
> >> > > 2. Create a build time switch that chooses between whatever the
> >> platform
> >> > > offers (ie the current arrangement) and stdcxx. Presumably
> >> defaulting
> >> to
> >> > > the
> >> > > current arrangement.
> >> > >
> >> > > I prefer option 2 but obviously it somewhat complicates our build
> >> > process
> >> > > and perhaps more seriously adds another complication to our test
> >> cases.
> >> > >
> >> > > What does the team think?
> >>
> >>
> >> My preference is also for option 2 as it gives our users more choice.
> >> > However, we may find ourselves #ifdef-ing chunks of code out to get
> >> around
> >> > the aforementioned differences in libraries (see Pete's map problem
> on
> >> > Windows yesterday..) which will make code less readable, etc. I think
> >> > starting this with the SDO codebase is a good idea, as this is a
> >> > relatively
> >> > standalone set of code, and will give us a good idea what the
> >> issues are
> >> > with this approach.
> >>
> >>
> >> Defintiely option 2!
> >>
> >> A side question - SDO has a couple of pre-reqs (libxml2, etc) - will
> >> these
> >> > need to be rebuilt against stdcxx as well?
> >>
> >>
> >> libxml2 is C rather than c++ so I don't think this is an issue. Same
> for
> >> Axis2C for the sdo_axiom utility.
> >>
> >> Cheers,
> >>
> >> --
> >> Pete
> >
> >
> >
> > Generall I +1 the move to stdcxx and specifically to option 2 because...
> >
> > In PHP SDO land, which of course uses Tuscany SDO, we tend to have faily
> > autonomous builds, i.e. we expect our users to be able to download the
> > SDO
> > package and build it without having to download too much other stuff. At
> > least other stuff that isn't already known to the PHP build system . We
> > already depend on libxml2, iconv & zlib but these are generally
> > available in
> > linux at stable versions. If we add a mandatory dependency on the apache
> > incubator version of stdxx then we have added another barrier to entry
> as
> > there is another packge to download that people may not be familiar
> > with and
> > may cause clashes with other extensions that use other solutions. We
> > won't
> > know until we try it. It may also cause problems for the automatic
> > windows
> > build that takes place on the PECL servers outside of our control. I
> > don't
> > really know the details of this though so would have to look into it.
> >
> > Simon
> >
>
> +1 for option 2 as well.
>
> --
> Jean-Sebastien
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: tuscany-dev-unsubscribe@ws.apache.org
> For additional commands, e-mail: tuscany-dev-help@ws.apache.org
>
>

Re: [SDO for C++] Opinions Requested: Adopting stdcxx as C++ library

Posted by Jean-Sebastien Delfino <js...@apache.org>.
Simon Laws wrote:
> On 8/30/06, Pete Robbins <ro...@googlemail.com> wrote:
>>
>> On 30/08/06, Andrew Borley <aj...@gmail.com> wrote:
>> >
>> > On 8/30/06, Geoffrey Winn <ge...@googlemail.com> wrote:
>> > <snip>. Using the Apache stdcxx library instead would provide
>> >
>> > > us with a number of benefits
>> >
>> >
>> > Agreed. +1 for this.
>>
>>
>> yup!
>>
>> The one difficulty is that once SDO links against the stdcxx library 
>> then
>> > > all users of SDO must also do so. I think this gives us two options
>> > >
>> > > 1. Just do it, and live with the consequences. In this case we will
>> (one
>> > > way
>> > > or another) pre-req stdcxx on all platforms, and all users of SDO 
>> for
>> > C++
>> > > will be required to use stdcxx as their C++ standard library.
>> > >
>> > > 2. Create a build time switch that chooses between whatever the
>> platform
>> > > offers (ie the current arrangement) and stdcxx. Presumably 
>> defaulting
>> to
>> > > the
>> > > current arrangement.
>> > >
>> > > I prefer option 2 but obviously it somewhat complicates our build
>> > process
>> > > and perhaps more seriously adds another complication to our test
>> cases.
>> > >
>> > > What does the team think?
>>
>>
>> My preference is also for option 2 as it gives our users more choice.
>> > However, we may find ourselves #ifdef-ing chunks of code out to get
>> around
>> > the aforementioned differences in libraries (see Pete's map problem on
>> > Windows yesterday..) which will make code less readable, etc. I think
>> > starting this with the SDO codebase is a good idea, as this is a
>> > relatively
>> > standalone set of code, and will give us a good idea what the 
>> issues are
>> > with this approach.
>>
>>
>> Defintiely option 2!
>>
>> A side question - SDO has a couple of pre-reqs (libxml2, etc) - will 
>> these
>> > need to be rebuilt against stdcxx as well?
>>
>>
>> libxml2 is C rather than c++ so I don't think this is an issue. Same for
>> Axis2C for the sdo_axiom utility.
>>
>> Cheers,
>>
>> -- 
>> Pete
>
>
>
> Generall I +1 the move to stdcxx and specifically to option 2 because...
>
> In PHP SDO land, which of course uses Tuscany SDO, we tend to have faily
> autonomous builds, i.e. we expect our users to be able to download the 
> SDO
> package and build it without having to download too much other stuff. At
> least other stuff that isn't already known to the PHP build system . We
> already depend on libxml2, iconv & zlib but these are generally 
> available in
> linux at stable versions. If we add a mandatory dependency on the apache
> incubator version of stdxx then we have added another barrier to entry as
> there is another packge to download that people may not be familiar 
> with and
> may cause clashes with other extensions that use other solutions. We 
> won't
> know until we try it. It may also cause problems for the automatic 
> windows
> build that takes place on the PECL servers outside of our control. I 
> don't
> really know the details of this though so would have to look into it.
>
> Simon
>

+1 for option 2 as well.

-- 
Jean-Sebastien


---------------------------------------------------------------------
To unsubscribe, e-mail: tuscany-dev-unsubscribe@ws.apache.org
For additional commands, e-mail: tuscany-dev-help@ws.apache.org


Re: [SDO for C++] Opinions Requested: Adopting stdcxx as C++ library

Posted by Simon Laws <si...@googlemail.com>.
On 8/30/06, Pete Robbins <ro...@googlemail.com> wrote:
>
> On 30/08/06, Andrew Borley <aj...@gmail.com> wrote:
> >
> > On 8/30/06, Geoffrey Winn <ge...@googlemail.com> wrote:
> > <snip>. Using the Apache stdcxx library instead would provide
> >
> > > us with a number of benefits
> >
> >
> > Agreed. +1 for this.
>
>
> yup!
>
> The one difficulty is that once SDO links against the stdcxx library then
> > > all users of SDO must also do so. I think this gives us two options
> > >
> > > 1. Just do it, and live with the consequences. In this case we will
> (one
> > > way
> > > or another) pre-req stdcxx on all platforms, and all users of SDO for
> > C++
> > > will be required to use stdcxx as their C++ standard library.
> > >
> > > 2. Create a build time switch that chooses between whatever the
> platform
> > > offers (ie the current arrangement) and stdcxx. Presumably defaulting
> to
> > > the
> > > current arrangement.
> > >
> > > I prefer option 2 but obviously it somewhat complicates our build
> > process
> > > and perhaps more seriously adds another complication to our test
> cases.
> > >
> > > What does the team think?
>
>
> My preference is also for option 2 as it gives our users more choice.
> > However, we may find ourselves #ifdef-ing chunks of code out to get
> around
> > the aforementioned differences in libraries (see Pete's map problem on
> > Windows yesterday..) which will make code less readable, etc. I think
> > starting this with the SDO codebase is a good idea, as this is a
> > relatively
> > standalone set of code, and will give us a good idea what the issues are
> > with this approach.
>
>
> Defintiely option 2!
>
> A side question - SDO has a couple of pre-reqs (libxml2, etc) - will these
> > need to be rebuilt against stdcxx as well?
>
>
> libxml2 is C rather than c++ so I don't think this is an issue. Same for
> Axis2C for the sdo_axiom utility.
>
> Cheers,
>
> --
> Pete



Generall I +1 the move to stdcxx and specifically to option 2 because...

In PHP SDO land, which of course uses Tuscany SDO, we tend to have faily
autonomous builds, i.e. we expect our users to be able to download the SDO
package and build it without having to download too much other stuff. At
least other stuff that isn't already known to the PHP build system . We
already depend on libxml2, iconv & zlib but these are generally available in
linux at stable versions. If we add a mandatory dependency on the apache
incubator version of stdxx then we have added another barrier to entry as
there is another packge to download that people may not be familiar with and
may cause clashes with other extensions that use other solutions. We won't
know until we try it. It may also cause problems for the automatic windows
build that takes place on the PECL servers outside of our control. I don't
really know the details of this though so would have to look into it.

Simon

Re: [SDO for C++] Opinions Requested: Adopting stdcxx as C++ library

Posted by Pete Robbins <ro...@googlemail.com>.
On 30/08/06, Andrew Borley <aj...@gmail.com> wrote:
>
> On 8/30/06, Geoffrey Winn <ge...@googlemail.com> wrote:
> <snip>. Using the Apache stdcxx library instead would provide
>
> > us with a number of benefits
>
>
> Agreed. +1 for this.


yup!

The one difficulty is that once SDO links against the stdcxx library then
> > all users of SDO must also do so. I think this gives us two options
> >
> > 1. Just do it, and live with the consequences. In this case we will (one
> > way
> > or another) pre-req stdcxx on all platforms, and all users of SDO for
> C++
> > will be required to use stdcxx as their C++ standard library.
> >
> > 2. Create a build time switch that chooses between whatever the platform
> > offers (ie the current arrangement) and stdcxx. Presumably defaulting to
> > the
> > current arrangement.
> >
> > I prefer option 2 but obviously it somewhat complicates our build
> process
> > and perhaps more seriously adds another complication to our test cases.
> >
> > What does the team think?


My preference is also for option 2 as it gives our users more choice.
> However, we may find ourselves #ifdef-ing chunks of code out to get around
> the aforementioned differences in libraries (see Pete's map problem on
> Windows yesterday..) which will make code less readable, etc. I think
> starting this with the SDO codebase is a good idea, as this is a
> relatively
> standalone set of code, and will give us a good idea what the issues are
> with this approach.


Defintiely option 2!

A side question - SDO has a couple of pre-reqs (libxml2, etc) - will these
> need to be rebuilt against stdcxx as well?


libxml2 is C rather than c++ so I don't think this is an issue. Same for
Axis2C for the sdo_axiom utility.

Cheers,

-- 
Pete

Re: [SDO for C++] Opinions Requested: Adopting stdcxx as C++ library

Posted by Andrew Borley <aj...@gmail.com>.
On 8/30/06, Geoffrey Winn <ge...@googlemail.com> wrote:
<snip>. Using the Apache stdcxx library instead would provide

> us with a number of benefits


Agreed. +1 for this.

The one difficulty is that once SDO links against the stdcxx library then
> all users of SDO must also do so. I think this gives us two options
>
> 1. Just do it, and live with the consequences. In this case we will (one
> way
> or another) pre-req stdcxx on all platforms, and all users of SDO for C++
> will be required to use stdcxx as their C++ standard library.
>
> 2. Create a build time switch that chooses between whatever the platform
> offers (ie the current arrangement) and stdcxx. Presumably defaulting to
> the
> current arrangement.
>
> I prefer option 2 but obviously it somewhat complicates our build process
> and perhaps more seriously adds another complication to our test cases.
>
> What does the team think?


My preference is also for option 2 as it gives our users more choice.
However, we may find ourselves #ifdef-ing chunks of code out to get around
the aforementioned differences in libraries (see Pete's map problem on
Windows yesterday..) which will make code less readable, etc. I think
starting this with the SDO codebase is a good idea, as this is a relatively
standalone set of code, and will give us a good idea what the issues are
with this approach.

A side question - SDO has a couple of pre-reqs (libxml2, etc) - will these
need to be rebuilt against stdcxx as well?

Cheers
Andy