Обсуждение: CORBA again. (was: light dawns: serious bug in FE/BE protocol handling)

Taral wrote:
> mico 2.2.6 has sufficient services support for CORBA to proceed. 

I know some people thinks mico is great but I have some concerns:
* I can't even get it to compile on my workstation at work, upgraded both g++ to egcs and libg++/libstd++. Still it
doesnot compile cleanly, and it is easy to track the problems in C++ (not my favoite) and strange #ifdefs. I think I
havegot some working version at home, will check.
 
* BIG! "make -k" takes forever. 
* Compiling the output stubs of MICO require a machine with at least
64MB RAM (hearsay). 
* requires C++ (and very specific versions too). This a new requirement for compiling PgSQL. (what state are the
C-bindingsin theses days?)
 

I am afraid that all this is going to put and end to the happy
"fetch src and compile on any favorite old platform" way of life
in the PgSQL-community. 

As long as these points are not resolved I think Tom L. will have to
continue fixing the old FE/BE-protocol, since CORBA will have to be
(very) optional on most platforms.

>[...]
> what I have now is not pretty). If anyone else wants to
> work with me on this, it might get done sooner. Otherwise, my research
> project takes precedence.

Why not put what you have in the CVS? The files in src/corba is pretty
old.
I have planned to get into this because did some coding around the 
FE/BE-protocol and would love to see a clean object-based communication
in PgSQL instead the current one. Let us see what you have got and what
is
still missing in more lightweight alternatives to MICO.
best regards,
Göran.

On Mon, 26 Apr 1999, Goran Thyni wrote:

> Taral wrote:
> > mico 2.2.6 has sufficient services support for CORBA to proceed. 
> 
> I know some people thinks mico is great but I have some concerns:

We are explicitly designing the system to allow any CORBA 2.2
implementation with the C++ mapping supported to be used.

Taral

On Mon, 26 Apr 1999, Goran Thyni wrote:

> Why not put what you have in the CVS? The files in src/corba is pretty
> old. I have planned to get into this because did some coding around
> the FE/BE-protocol and would love to see a clean object-based
> communication in PgSQL instead the current one. Let us see what you
> have got and what is still missing in more lightweight alternatives to
> MICO.

What I have is not worth putting in the CVS, plus it is not finalized, nor
is it working in any real way. (It's just a set of IDL files, attached.)

The most important thing we need support for is CORBA 2.2 itself, which
when I started this, was only supported by CORBA. Also, I would like
support for the LifeCycle service (I think), the Security service, the
Transaction service, and maybe others I can't think of right now.

Taral

On Mon, 26 Apr 1999, Taral wrote:

> We are explicitly designing the system to allow any CORBA 2.2
> implementation with the C++ mapping supported to be used.

Given that PostgreSQL is a C project, why not use a C ORB like ORBit?
It's a whole lot faster than MICO...

--
Todd Graham Lewis                        Postmaster, MindSpring Enterprises
tlewis@mindspring.net                                (800) 719-4664, x22804

"A pint of sweat will save a gallon of blood."          -- George S. Patton

On Mon, 26 Apr 1999, Taral wrote:

> The most important thing we need support for is CORBA 2.2 itself, which
> when I started this, was only supported by CORBA. Also, I would like
> support for the LifeCycle service (I think), the Security service, the
> Transaction service, and maybe others I can't think of right now.

I am almost done with the first version of the GNU Transaction Server.
It uses ORBit for its own needs, but there's no reason you couldn't use
other ORBs as clients if they were appropriately modified.  I plan on
so modifying ORBit.

--
Todd Graham Lewis                        Postmaster, MindSpring Enterprises
tlewis@mindspring.net                                (800) 719-4664, x22804

"A pint of sweat will save a gallon of blood."          -- George S. Patton

On Mon, 26 Apr 1999, Taral wrote:

> What I have is not worth putting in the CVS, plus it is not finalized, nor
> is it working in any real way. (It's just a set of IDL files, attached.)

Helps if I attach them.

Taral

On Mon, 26 Apr 1999, Todd Graham Lewis wrote:

> Given that PostgreSQL is a C project, why not use a C ORB like ORBit?
> It's a whole lot faster than MICO...

Does it fully support 2.2?

Taral

On Mon, 26 Apr 1999, Todd Graham Lewis wrote:

> I am almost done with the first version of the GNU Transaction Server.
> It uses ORBit for its own needs, but there's no reason you couldn't use
> other ORBs as clients if they were appropriately modified.  I plan on
> so modifying ORBit.

PostgreSQL uses a Berkeley license. That's one of the (many) reasons we
don't want to bundle any ORB into the package.

Do we still have problems depending on GPL code?

Taral

>>>>> "t" == Taral  <taral@taral.net>
>>>>> wrote the following on Mon, 26 Apr 1999 13:20:55 -0500 (CDT)

t> On Mon, 26 Apr 1999, Todd Graham Lewis wrote:
>> Given that PostgreSQL is a C project, why not use a C ORB like ORBit?
>> It's a whole lot faster than MICO...

t> Does it fully support 2.2?

yes. And i've written a multi database package for the Gnome Project
using ORbit. Take a look at <http://www.lausch.at/gda> to find the
documentation. You'll get the source from the gnome ftp server. It's
called gnome-db and postgres is already a supported database.,

t> Taral
--
Michael Lausch
See my web page <http://www.lausch.at/> or query PGP key server for PGP key.
"Reality is that which, when you stop believing in it, doesn't go away".               -- Philip K. Dick

On Mon, 26 Apr 1999, Taral wrote:

> PostgreSQL uses a Berkeley license. That's one of the (many) reasons we
> don't want to bundle any ORB into the package.
> 
> Do we still have problems depending on GPL code?

ORBit is LGPL'd, so there's no license conflict.  And you'd just be
linking against it, not really "including" it, depending on how formal
you want to be about it.  The last time we talked about this, I think
it was announced that ORBit would be acceptable, but that was a while
ago.  Are there searchable archives of the Postgres list?

--
Todd Graham Lewis                        Postmaster, MindSpring Enterprises
tlewis@mindspring.net                                (800) 719-4664, x22804

"A pint of sweat will save a gallon of blood."          -- George S. Patton

I use KDE at home, as well as at work...KDE uses MICO...

The CORBA implemetaion, as discussed before, is going to have to be
'wrappered' so that it isn't tied to any particular implementation...

Taral was, I believe, planning on starting with MICO and then let's those
using other implementations build up from there...


On Mon, 26 Apr 1999, Michael Lausch wrote:

> >>>>> "t" == Taral  <taral@taral.net>
> >>>>> wrote the following on Mon, 26 Apr 1999 13:20:55 -0500 (CDT)
> 
> t> On Mon, 26 Apr 1999, Todd Graham Lewis wrote:
> >> Given that PostgreSQL is a C project, why not use a C ORB like ORBit?
> >> It's a whole lot faster than MICO...
> 
> t> Does it fully support 2.2?
> 
> yes. And i've written a multi database package for the Gnome Project
> using ORbit. Take a look at <http://www.lausch.at/gda> to find the
> documentation. You'll get the source from the gnome ftp server. It's
> called gnome-db and postgres is already a supported database.,
> 
> t> Taral
> --
> Michael Lausch
> See my web page <http://www.lausch.at/> or query PGP key server for PGP key.
> "Reality is that which, when you stop believing in it, doesn't go away".
>                 -- Philip K. Dick
> 

Marc G. Fournier                   ICQ#7615664               IRC Nick: Scrappy
Systems Administrator @ hub.org 
primary: scrappy@hub.org           secondary: scrappy@{freebsd|postgresql}.org 

Well, I don't mind the idea of using ORBit, since it _is_ smaller. I'm
looking into it now. However, I do intend to wrapper the whole lot. The
only thing is to pick a required mapping, and I said (way back then) that
I preferred C since it had substantially better handling of unions and the
like.

Taral

On Mon, 26 Apr 1999, The Hermit Hacker wrote:

> 
> I use KDE at home, as well as at work...KDE uses MICO...
> 
> The CORBA implemetaion, as discussed before, is going to have to be
> 'wrappered' so that it isn't tied to any particular implementation...
> 
> Taral was, I believe, planning on starting with MICO and then let's those
> using other implementations build up from there...
> 
> 
> On Mon, 26 Apr 1999, Michael Lausch wrote:
> 
> > >>>>> "t" == Taral  <taral@taral.net>
> > >>>>> wrote the following on Mon, 26 Apr 1999 13:20:55 -0500 (CDT)
> > 
> > t> On Mon, 26 Apr 1999, Todd Graham Lewis wrote:
> > >> Given that PostgreSQL is a C project, why not use a C ORB like ORBit?
> > >> It's a whole lot faster than MICO...
> > 
> > t> Does it fully support 2.2?
> > 
> > yes. And i've written a multi database package for the Gnome Project
> > using ORbit. Take a look at <http://www.lausch.at/gda> to find the
> > documentation. You'll get the source from the gnome ftp server. It's
> > called gnome-db and postgres is already a supported database.,
> > 
> > t> Taral
> > --
> > Michael Lausch
> > See my web page <http://www.lausch.at/> or query PGP key server for PGP key.
> > "Reality is that which, when you stop believing in it, doesn't go away".
> >                 -- Philip K. Dick
> > 
> 
> Marc G. Fournier                   ICQ#7615664               IRC Nick: Scrappy
> Systems Administrator @ hub.org 
> primary: scrappy@hub.org           secondary: scrappy@{freebsd|postgresql}.org 
> 

On Mon, 26 Apr 1999, The Hermit Hacker wrote:

> I use KDE at home, as well as at work...KDE uses MICO...
> 
> The CORBA implemetaion, as discussed before, is going to have to be
> 'wrappered' so that it isn't tied to any particular implementation...

The language bindings for CORBA are completely standardized.  The only
thing you have to change are your header files and link options.

That said, the CORBA C bindings are different from the CORBA C++ bindings,
and so you are going to have to chose which language to pursue.  I suggest
that staying with C is the natural choice for PostgreSQL, and you have
two very good free ORBs to chose from: ORBit and ILU.  For C++, you
have one more real choice: OmniORB, MICO, and TAO.  The two performance
champs are ORBit and TAO, which fortunately enough covers both of your
language choices.  The language issue weighs heavily in my mind as a C
programmer who doesn't want to play the popular-subset- of-the-month
game with C++.  (I think it's templates this month, although virtual
somethings are making a comeback.  But don't use multiple inheritance!
That was all the rage back in '95, but now only dorks use it!)

Plus, if you code the Postgres part with ORBit, there's no reason why
clients can't use MICO; that's the whole point.

> Taral was, I believe, planning on starting with MICO and then let's those
> using other implementations build up from there...

Is all of the underlying glue work going to be done in C++?

--
Todd Graham Lewis                        Postmaster, MindSpring Enterprises
tlewis@mindspring.net                                (800) 719-4664, x22804

"A pint of sweat will save a gallon of blood."          -- George S. Patton

(As I look at this again)

On Mon, 26 Apr 1999, Michael Lausch wrote:

> yes. And i've written a multi database package for the Gnome Project
> using ORbit. Take a look at <http://www.lausch.at/gda> to find the
> documentation. You'll get the source from the gnome ftp server. It's
> called gnome-db and postgres is already a supported database.,

Ahh.. Yeah, I intend to map the postgresql functionality onto COSS as much
as possible, so I'm implementing the Query Service.

Taral