From boris at codesynthesis.com  Fri Nov  1 11:06:39 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Fri Nov  1 11:07:27 2019
Subject: [odb-users] How to cache the queried object pointer?
In-Reply-To: <MN2PR12MB30247C53AC737CBBDC7ABB53CB630@MN2PR12MB3024.namprd12.prod.outlook.com>
References: <MN2PR12MB30247C53AC737CBBDC7ABB53CB630@MN2PR12MB3024.namprd12.prod.outlook.com>
Message-ID: <boris.20191101170424@codesynthesis.com>

Justin Huang <yohuang@nvidia.com> writes:

>            typedef odb::result<Module> result;
>            result all_modules (db->query<Module> ());
> 
>            int count = 0;
>            for (result::iterator i(all_modules.begin()); i != all_modules.end(); ++i) {
>                 auto ptr = i.operator->();
>                 modules.push_back(ptr);

You should use result::iterator::load() (i.load()) to load an object from
a query result. See Section 4.4, "Query Result" for details.

From boris at codesynthesis.com  Fri Nov  1 11:12:36 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Fri Nov  1 11:13:25 2019
Subject: [odb-users] Is that possible to order the child objects based on
	some specific column
In-Reply-To: <tencent_6ED00839A4F2FD876174F158EF510D96110A@qq.com>
References: <tencent_6ED00839A4F2FD876174F158EF510D96110A@qq.com>
Message-ID: <boris.20191101170734@codesynthesis.com>

???? <icewill9999@vip.qq.com> writes:

> Say i have two tables, A and B of 1:N relationship, B is eager loaded.
> 
> 
> #pragma db object
> class A
> {
> ........
>   vector<weak_ptr<B>> b_;
> 
> }
> 
> 
> class B
> {
> .......
> int column_C;
> }
> 
> 
> i want the B objects was ordered by column_C in the vector. Is that
> possible to do that? i use odb 2.4.0.

There is no built-in support at the ODB level for this. You can check
if it's possible at the database level for the database that you use
(i.e., some kind of a default sort order). Alternatively, you can use
an ODB view (probably an object-loading view) with an ORDER BY clause
to manually load the vector (say from a database callback).

From sean.clarke at sec-consulting.co.uk  Sat Nov  2 12:27:24 2019
From: sean.clarke at sec-consulting.co.uk (Sean Clarke)
Date: Sat Nov  2 12:28:33 2019
Subject: [odb-users] ETA for ODB 2.5.0 (RC or release) packages
Message-ID: <CAHCKNJgV9HzsDPWByLzria1F3N6Us8xE7kSKaz-P69=pTdCXVw@mail.gmail.com>

Hi,
    is there a timeline for when we can expect some official distro
packages for ODB 2.5.0? - it seems to have been in development and beta for
a very long time. I have previously battled with build2 and it produced
some working output, however these could not be readily used on a
vanilla Debian Buster or Ubuntu platform as build2 pulled in its
dependencies at different revisions from the distro platforms.

Regards
Sean
From per.edin at sequence-point.se  Sat Nov  2 15:12:44 2019
From: per.edin at sequence-point.se (Per Edin)
Date: Mon Nov  4 09:10:26 2019
Subject: [odb-users] ETA for ODB 2.5.0 (RC or release) packages
In-Reply-To: <CAHCKNJgV9HzsDPWByLzria1F3N6Us8xE7kSKaz-P69=pTdCXVw@mail.gmail.com>
References: <CAHCKNJgV9HzsDPWByLzria1F3N6Us8xE7kSKaz-P69=pTdCXVw@mail.gmail.com>
Message-ID: <CAJA0JPAZXcC2rE0tqnK-nTOcogjuFi1Te0Dg9_CFom1b-7Mgnw@mail.gmail.com>

Hi!

On Sat, Nov 2, 2019 at 5:28 PM Sean Clarke
<sean.clarke@sec-consulting.co.uk> wrote:
> I have previously battled with build2 and it produced
> some working output, however these could not be readily used on a
> vanilla Debian Buster or Ubuntu platform as build2 pulled in its
> dependencies at different revisions from the distro platforms.
>

I can't say anything about when 2.5.0 will be released, but I'm
curious about the problems you're having with build2.

I'm myself using ODB and build2 on multiple machines (Ubuntu, Debian
and macOS) and I haven't yet experienced any problems related to
system installed packages.

// Per

From boris at codesynthesis.com  Tue Nov  5 09:28:56 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Tue Nov  5 09:29:59 2019
Subject: [odb-users] ETA for ODB 2.5.0 (RC or release) packages
In-Reply-To: <CAHCKNJgV9HzsDPWByLzria1F3N6Us8xE7kSKaz-P69=pTdCXVw@mail.gmail.com>
References: <CAHCKNJgV9HzsDPWByLzria1F3N6Us8xE7kSKaz-P69=pTdCXVw@mail.gmail.com>
Message-ID: <boris.20191105155720@codesynthesis.com>

Sean Clarke <sean.clarke@sec-consulting.co.uk> writes:

>     is there a timeline for when we can expect some official distro
> packages for ODB 2.5.0?

Uh, I don't think anyone can predict when (or even if) the next release
of ODB (once released) will appear in official distribution package
repositories -- this is done by volunteers and depends on each
distribution's policies/priorities/etc.

As an example, there was bug 1588330[1] in an Ubuntu LTS release
where libodb and the rest of the database runtime libraries were
compiled with different versions of GCC which happened to have
different ABIs (so the fix was to just rebuild all the ODB packages
with the same version of GCC). Guess how long it took to fix? Almost
a year (reported on 2016-06-02, fixed on 2017-02-09).

And during that time there would be an email on odb-users every
week or two complaining about the issue. And I had to keep replying,
and explaining, and telling people there is nothing I can do until
Ubuntu decides this problem is worthy of their attention.

So we started build2 where everything builds exactly the same no
matter the platform (even on Windows) and where we actually have
the ability to fix things when they break.

I am not denying that using build2 is probably less convenient than
using your distribution's package manager, at least initially. But
it's a tradeoff: in return we can fix issues and make them available
to you in a matter of minutes.


> [...] - it seems to have been in development and beta for
> a very long time.

The ODB itself is fairly mature and at this stage of the 2.5.0 cycle
we mostly fix bugs. So the code itself could be released tomorrow,
except that the build system support is not finished yet. Specifically,
we need to finish two major things:

1. Implement the ODB code generation module for build2.

2. With that support, port odb-tests packages to build2.

The good news is in the upcoming release of build2 we have support
for build system modules so we can make progress on (1).



> I have previously battled with build2 and it produced some
> working output, however these could not be readily used on a
> vanilla Debian Buster or Ubuntu platform as build2 pulled in its
> dependencies at different revisions from the distro platforms.

I would be happy to help you with this if you can provide more
details. In particular, you can instruct build2 to use system-
installed libraries for any of the dependencies. The installation
instructions[2] mention this explicitly:

"Note that by default the underlying database libraries (libsqlite3,
libmysqlclient, and libpq) will be built from source packages. If you would
prefer to use the system-installed versions, adjust the corresponding build
commands as follows:"

$ bpkg build libodb-sqlite ?sys:libsqlite3
$ bpkg build libodb-pgsql  ?sys:libpq
$ bpkg build libodb-mysql  ?sys:libmysqlclient

[1] https://bugs.launchpad.net/ubuntu/+source/libodb/+bug/1588330
[2] https://codesynthesis.com/products/odb/doc/install-build2.xhtml

From yohuang at nvidia.com  Mon Nov  4 21:47:13 2019
From: yohuang at nvidia.com (Justin Huang)
Date: Tue Nov  5 09:36:08 2019
Subject: [odb-users] How to cache the queried object pointer?
In-Reply-To: <boris.20191101170424@codesynthesis.com>
References: <MN2PR12MB30247C53AC737CBBDC7ABB53CB630@MN2PR12MB3024.namprd12.prod.outlook.com>
	<boris.20191101170424@codesynthesis.com>
Message-ID: <MN2PR12MB302499F1D037634CA5C9C4A0CB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>

Hi Boris,

Thanks. It works.

But I saw a weird issue when I use i.load() to query data members.

Below I have a Module class, Instance class and Port class. I persisted the ports() and instances() of Module object, and confirmed that they were persisted in sqlite3 DB.

I tried to use i.load() pointer to access ports(), the returned ports() size is 0, but size of instances() is not 0.

        session s;
        db = std::unique_ptr<odb::database>(create_database (argc, argv));

        // load modules
        {
            odb::transaction t (db->begin());

            typedef odb::result<Module> result;
            result all_modules (db->query<Module> ());

            int count = 0;
            for (result::iterator i(all_modules.begin()); i != all_modules.end(); ++i) {
                std::shared_ptr<Module> p (i.load());
                std::cout << "loading module: " << p->name() << std::endl;
                std::cout << "ports size: "            << p->ports().size() << std::endl;
                std::cout << "instances size: "    << p->instances().size() << std::endl;
                for (auto port: p->ports())
                    std::cout << "loading module port: " << port->name() << std::endl;
             
                for (auto inst: p->instances())
                    std::cout << "loading module inst: " << inst->name() << std::endl;
            }

Any suggestions to debug this?

Thanks.

-------------------------------------------------------
# Module, Port, Instance class declaration
-------------------------------------------------------
#pragma db object
class Module: public NamedHessObject
{
public:
    Module(std::string name);
   ~Module();

    std::vector<Port*>&               ports();
    std::vector<Instance*>&       instances();
    std::vector<Connection*>& connections();
    std::string                                as_string();
    static Module*                       find_by_name(std::string name);

public:
    #pragma db transient
    static std::map<std::string, Module*> modules_;
    static size_t count;

private:
    friend class odb::access;
    Module();

private:
    std::vector<Port*>               ports_;
    std::vector<Instance*>       instances_;
    std::vector<Connection*> connections_;
}; // class Module

#pragma db object
class Instance: public NamedHessObject
{
public:
    Instance(std::string name);
   ~Instance();

    Module*                 module();
    void                         module(Module* module);

    Module*                 owner();
    void                         owner(Module* owner);

    std::vector<PortRef *>& portrefs();
    std::string                          as_string();

public:
    static size_t count;

private:
    friend class odb::access;
    Instance();

private:
    Module*                           owner_;  // owner module
    Module*                           module_; // instantiated module
    std::vector<PortRef *>  portrefs_;
}; // class Instance

#pragma db object
class Port: public NamedHessObject
{
public:
    Port(std::string name);
    Port();
   ~Port();

    std::string                direction();
    void                           direction(std::string direction);

    Module*                    owner();
    void                             owner(Module* owner);

    Interface*                 interface();
    void                            interface(Interface* interface);

    std::vector<Connection *>& connections();
    std::string                                  as_string();

public:
    static size_t count;

private:
    friend class odb::access;

private:
    std::string                 direction_;
    Module*                   owner_;       // module this port belongs to.
    Interface*                interface_;   // instantiated interface.
    std::vector<Connection *> connections_; // channels connected to this port inside the owner_ module
}; // class Port



-----Original Message-----
From: Boris Kolpackov <boris@codesynthesis.com> 
Sent: Friday, November 1, 2019 11:07 PM
To: Justin Huang <yohuang@nvidia.com>
Cc: odb-users@codesynthesis.com
Subject: Re: [odb-users] How to cache the queried object pointer?

Justin Huang <yohuang@nvidia.com> writes:

>            typedef odb::result<Module> result;
>            result all_modules (db->query<Module> ());
> 
>            int count = 0;
>            for (result::iterator i(all_modules.begin()); i != all_modules.end(); ++i) {
>                 auto ptr = i.operator->();
>                 modules.push_back(ptr);

You should use result::iterator::load() (i.load()) to load an object from a query result. See Section 4.4, "Query Result" for details.
-----------------------------------------------------------------------------------
This email message is for the sole use of the intended recipient(s) and may contain
confidential information.  Any unauthorized review, use, disclosure or distribution
is prohibited.  If you are not the intended recipient, please contact the sender by
reply email and destroy all copies of the original message.
-----------------------------------------------------------------------------------

From yohuang at nvidia.com  Mon Nov  4 23:56:24 2019
From: yohuang at nvidia.com (Justin Huang)
Date: Tue Nov  5 09:36:08 2019
Subject: [odb-users] How to cache the queried object pointer?
In-Reply-To: <MN2PR12MB302499F1D037634CA5C9C4A0CB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>
References: <MN2PR12MB30247C53AC737CBBDC7ABB53CB630@MN2PR12MB3024.namprd12.prod.outlook.com>
	<boris.20191101170424@codesynthesis.com>
	<MN2PR12MB302499F1D037634CA5C9C4A0CB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>
Message-ID: <MN2PR12MB3024F812D1CFE299328F2CE5CB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>

It is weird that there was no Module_ports table when I loaded the sqlite3 DB.



-----Original Message-----
From: Justin Huang 
Sent: Tuesday, November 5, 2019 10:47 AM
To: odb-users@codesynthesis.com
Subject: RE: [odb-users] How to cache the queried object pointer?

Hi Boris,

Thanks. It works.

But I saw a weird issue when I use i.load() to query data members.

Below I have a Module class, Instance class and Port class. I persisted the ports() and instances() of Module object, and confirmed that they were persisted in sqlite3 DB.

I tried to use i.load() pointer to access ports(), the returned ports() size is 0, but size of instances() is not 0.

        session s;
        db = std::unique_ptr<odb::database>(create_database (argc, argv));

        // load modules
        {
            odb::transaction t (db->begin());

            typedef odb::result<Module> result;
            result all_modules (db->query<Module> ());

            int count = 0;
            for (result::iterator i(all_modules.begin()); i != all_modules.end(); ++i) {
                std::shared_ptr<Module> p (i.load());
                std::cout << "loading module: " << p->name() << std::endl;
                std::cout << "ports size: "            << p->ports().size() << std::endl;
                std::cout << "instances size: "    << p->instances().size() << std::endl;
                for (auto port: p->ports())
                    std::cout << "loading module port: " << port->name() << std::endl;
             
                for (auto inst: p->instances())
                    std::cout << "loading module inst: " << inst->name() << std::endl;
            }

Any suggestions to debug this?

Thanks.

-------------------------------------------------------
# Module, Port, Instance class declaration
-------------------------------------------------------
#pragma db object
class Module: public NamedHessObject
{
public:
    Module(std::string name);
   ~Module();

    std::vector<Port*>&               ports();
    std::vector<Instance*>&       instances();
    std::vector<Connection*>& connections();
    std::string                                as_string();
    static Module*                       find_by_name(std::string name);

public:
    #pragma db transient
    static std::map<std::string, Module*> modules_;
    static size_t count;

private:
    friend class odb::access;
    Module();

private:
    std::vector<Port*>               ports_;
    std::vector<Instance*>       instances_;
    std::vector<Connection*> connections_; }; // class Module

#pragma db object
class Instance: public NamedHessObject
{
public:
    Instance(std::string name);
   ~Instance();

    Module*                 module();
    void                         module(Module* module);

    Module*                 owner();
    void                         owner(Module* owner);

    std::vector<PortRef *>& portrefs();
    std::string                          as_string();

public:
    static size_t count;

private:
    friend class odb::access;
    Instance();

private:
    Module*                           owner_;  // owner module
    Module*                           module_; // instantiated module
    std::vector<PortRef *>  portrefs_;
}; // class Instance

#pragma db object
class Port: public NamedHessObject
{
public:
    Port(std::string name);
    Port();
   ~Port();

    std::string                direction();
    void                           direction(std::string direction);

    Module*                    owner();
    void                             owner(Module* owner);

    Interface*                 interface();
    void                            interface(Interface* interface);

    std::vector<Connection *>& connections();
    std::string                                  as_string();

public:
    static size_t count;

private:
    friend class odb::access;

private:
    std::string                 direction_;
    Module*                   owner_;       // module this port belongs to.
    Interface*                interface_;   // instantiated interface.
    std::vector<Connection *> connections_; // channels connected to this port inside the owner_ module }; // class Port



-----Original Message-----
From: Boris Kolpackov <boris@codesynthesis.com>
Sent: Friday, November 1, 2019 11:07 PM
To: Justin Huang <yohuang@nvidia.com>
Cc: odb-users@codesynthesis.com
Subject: Re: [odb-users] How to cache the queried object pointer?

Justin Huang <yohuang@nvidia.com> writes:

>            typedef odb::result<Module> result;
>            result all_modules (db->query<Module> ());
> 
>            int count = 0;
>            for (result::iterator i(all_modules.begin()); i != all_modules.end(); ++i) {
>                 auto ptr = i.operator->();
>                 modules.push_back(ptr);

You should use result::iterator::load() (i.load()) to load an object from a query result. See Section 4.4, "Query Result" for details.
-----------------------------------------------------------------------------------
This email message is for the sole use of the intended recipient(s) and may contain
confidential information.  Any unauthorized review, use, disclosure or distribution
is prohibited.  If you are not the intended recipient, please contact the sender by
reply email and destroy all copies of the original message.
-----------------------------------------------------------------------------------

From yohuang at nvidia.com  Tue Nov  5 00:25:40 2019
From: yohuang at nvidia.com (Justin Huang)
Date: Tue Nov  5 09:36:09 2019
Subject: [odb-users] How to cache the queried object pointer?
In-Reply-To: <MN2PR12MB3024F812D1CFE299328F2CE5CB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>
References: <MN2PR12MB30247C53AC737CBBDC7ABB53CB630@MN2PR12MB3024.namprd12.prod.outlook.com>
	<boris.20191101170424@codesynthesis.com>
	<MN2PR12MB302499F1D037634CA5C9C4A0CB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>
	<MN2PR12MB3024F812D1CFE299328F2CE5CB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>
Message-ID: <MN2PR12MB3024E589C41ED31A6709A0DDCB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>

It is quite weird that if I reorder the declaration of ports_, instances_ and connections_ data members of Module class, ODB would always do not create table for the first one. In below example, there is no 'CREATE TABLE' statement for Modulel_ports. If I move instances_ up above ports_, then ODB would not generate CREATE TABLE statement for Module_instances. 

I can't understand what is going wrong here.


class Module .. {
private:
    std::vector<Port*>               ports_;
    std::vector<Instance*>       instances_;
    std::vector<Connection*> connections_;
}; // class Module

-----Original Message-----
From: Justin Huang 
Sent: Tuesday, November 5, 2019 12:56 PM
To: odb-users@codesynthesis.com
Subject: RE: [odb-users] How to cache the queried object pointer?

It is weird that there was no Module_ports table when I loaded the sqlite3 DB.



-----Original Message-----
From: Justin Huang 
Sent: Tuesday, November 5, 2019 10:47 AM
To: odb-users@codesynthesis.com
Subject: RE: [odb-users] How to cache the queried object pointer?

Hi Boris,

Thanks. It works.

But I saw a weird issue when I use i.load() to query data members.

Below I have a Module class, Instance class and Port class. I persisted the ports() and instances() of Module object, and confirmed that they were persisted in sqlite3 DB.

I tried to use i.load() pointer to access ports(), the returned ports() size is 0, but size of instances() is not 0.

        session s;
        db = std::unique_ptr<odb::database>(create_database (argc, argv));

        // load modules
        {
            odb::transaction t (db->begin());

            typedef odb::result<Module> result;
            result all_modules (db->query<Module> ());

            int count = 0;
            for (result::iterator i(all_modules.begin()); i != all_modules.end(); ++i) {
                std::shared_ptr<Module> p (i.load());
                std::cout << "loading module: " << p->name() << std::endl;
                std::cout << "ports size: "            << p->ports().size() << std::endl;
                std::cout << "instances size: "    << p->instances().size() << std::endl;
                for (auto port: p->ports())
                    std::cout << "loading module port: " << port->name() << std::endl;
             
                for (auto inst: p->instances())
                    std::cout << "loading module inst: " << inst->name() << std::endl;
            }

Any suggestions to debug this?

Thanks.

-------------------------------------------------------
# Module, Port, Instance class declaration
-------------------------------------------------------
#pragma db object
class Module: public NamedHessObject
{
public:
    Module(std::string name);
   ~Module();

    std::vector<Port*>&               ports();
    std::vector<Instance*>&       instances();
    std::vector<Connection*>& connections();
    std::string                                as_string();
    static Module*                       find_by_name(std::string name);

public:
    #pragma db transient
    static std::map<std::string, Module*> modules_;
    static size_t count;

private:
    friend class odb::access;
    Module();

private:
    std::vector<Port*>               ports_;
    std::vector<Instance*>       instances_;
    std::vector<Connection*> connections_; }; // class Module

#pragma db object
class Instance: public NamedHessObject
{
public:
    Instance(std::string name);
   ~Instance();

    Module*                 module();
    void                         module(Module* module);

    Module*                 owner();
    void                         owner(Module* owner);

    std::vector<PortRef *>& portrefs();
    std::string                          as_string();

public:
    static size_t count;

private:
    friend class odb::access;
    Instance();

private:
    Module*                           owner_;  // owner module
    Module*                           module_; // instantiated module
    std::vector<PortRef *>  portrefs_;
}; // class Instance

#pragma db object
class Port: public NamedHessObject
{
public:
    Port(std::string name);
    Port();
   ~Port();

    std::string                direction();
    void                           direction(std::string direction);

    Module*                    owner();
    void                             owner(Module* owner);

    Interface*                 interface();
    void                            interface(Interface* interface);

    std::vector<Connection *>& connections();
    std::string                                  as_string();

public:
    static size_t count;

private:
    friend class odb::access;

private:
    std::string                 direction_;
    Module*                   owner_;       // module this port belongs to.
    Interface*                interface_;   // instantiated interface.
    std::vector<Connection *> connections_; // channels connected to this port inside the owner_ module }; // class Port



-----Original Message-----
From: Boris Kolpackov <boris@codesynthesis.com>
Sent: Friday, November 1, 2019 11:07 PM
To: Justin Huang <yohuang@nvidia.com>
Cc: odb-users@codesynthesis.com
Subject: Re: [odb-users] How to cache the queried object pointer?

Justin Huang <yohuang@nvidia.com> writes:

>            typedef odb::result<Module> result;
>            result all_modules (db->query<Module> ());
> 
>            int count = 0;
>            for (result::iterator i(all_modules.begin()); i != all_modules.end(); ++i) {
>                 auto ptr = i.operator->();
>                 modules.push_back(ptr);

You should use result::iterator::load() (i.load()) to load an object from a query result. See Section 4.4, "Query Result" for details.
-----------------------------------------------------------------------------------
This email message is for the sole use of the intended recipient(s) and may contain
confidential information.  Any unauthorized review, use, disclosure or distribution
is prohibited.  If you are not the intended recipient, please contact the sender by
reply email and destroy all copies of the original message.
-----------------------------------------------------------------------------------

From boris at codesynthesis.com  Tue Nov  5 09:38:12 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Tue Nov  5 09:39:13 2019
Subject: [odb-users] How to cache the queried object pointer?
In-Reply-To: <MN2PR12MB302499F1D037634CA5C9C4A0CB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>
References: <MN2PR12MB30247C53AC737CBBDC7ABB53CB630@MN2PR12MB3024.namprd12.prod.outlook.com>
	<boris.20191101170424@codesynthesis.com>
	<MN2PR12MB302499F1D037634CA5C9C4A0CB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>
Message-ID: <boris.20191105163623@codesynthesis.com>

Justin Huang <yohuang@nvidia.com> writes:

> public:
>     #pragma db transient
>     static std::map<std::string, Module*> modules_;
>     static size_t count;
>
> private:
>     std::vector<Port*>               ports_;
>     std::vector<Instance*>       instances_;
>     std::vector<Connection*> connections_;

You don't need to mark static data members as transient. And I think
what happens here is that db transient pragma applies to the first
non-static data member, which is ports_.

From yohuang at nvidia.com  Tue Nov  5 10:15:42 2019
From: yohuang at nvidia.com (Justin Huang)
Date: Wed Nov  6 06:30:18 2019
Subject: [odb-users] How to cache the queried object pointer?
In-Reply-To: <boris.20191105163623@codesynthesis.com>
References: <MN2PR12MB30247C53AC737CBBDC7ABB53CB630@MN2PR12MB3024.namprd12.prod.outlook.com>
	<boris.20191101170424@codesynthesis.com>
	<MN2PR12MB302499F1D037634CA5C9C4A0CB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>
	<boris.20191105163623@codesynthesis.com>
Message-ID: <MN2PR12MB3024FDFA6E8CC760A65301D8CB7E0@MN2PR12MB3024.namprd12.prod.outlook.com>

Hi Boris,

You root caused it so quickly. Thanks very much! The Module_ports table is now created when I commented out #pragma db transient. I spent a lot of time to find a workaround, i.e., I added a dummy data member before ports_. Now I can remove it.

Maybe this can be documented in the manual html page.

Thanks,
Justin

-----Original Message-----
From: Boris Kolpackov <boris@codesynthesis.com> 
Sent: Tuesday, November 5, 2019 10:38 PM
To: Justin Huang <yohuang@nvidia.com>
Cc: odb-users@codesynthesis.com
Subject: Re: [odb-users] How to cache the queried object pointer?

Justin Huang <yohuang@nvidia.com> writes:

> public:
>     #pragma db transient
>     static std::map<std::string, Module*> modules_;
>     static size_t count;
>
> private:
>     std::vector<Port*>               ports_;
>     std::vector<Instance*>       instances_;
>     std::vector<Connection*> connections_;

You don't need to mark static data members as transient. And I think what happens here is that db transient pragma applies to the first non-static data member, which is ports_.
-----------------------------------------------------------------------------------
This email message is for the sole use of the intended recipient(s) and may contain
confidential information.  Any unauthorized review, use, disclosure or distribution
is prohibited.  If you are not the intended recipient, please contact the sender by
reply email and destroy all copies of the original message.
-----------------------------------------------------------------------------------

From sean.clarke at sec-consulting.co.uk  Fri Nov  8 01:56:05 2019
From: sean.clarke at sec-consulting.co.uk (Sean Clarke)
Date: Fri Nov  8 01:57:37 2019
Subject: [odb-users] ETA for ODB 2.5.0 (RC or release) packages
In-Reply-To: <boris.20191105155720@codesynthesis.com>
References: <CAHCKNJgV9HzsDPWByLzria1F3N6Us8xE7kSKaz-P69=pTdCXVw@mail.gmail.com>
	<boris.20191105155720@codesynthesis.com>
Message-ID: <CAHCKNJgFYznpmRVehk6U2y=DL2oPp+qM5mFr=-UjBALPOgqWPQ@mail.gmail.com>

On Tue, 5 Nov 2019 at 14:29, Boris Kolpackov <boris@codesynthesis.com>
wrote:

> Sean Clarke <sean.clarke@sec-consulting.co.uk> writes:
>
> >     is there a timeline for when we can expect some official distro
> > packages for ODB 2.5.0?
>
> Uh, I don't think anyone can predict when (or even if) the next release
> of ODB (once released) will appear in official distribution package
> repositories -- this is done by volunteers and depends on each
> distribution's policies/priorities/etc.
>


Sorry, I should have been clearer - I was referring to the debian packages
available on your website.. I gave up on the distros distributed version
long ago for the reasons you give.



> I am not denying that using build2 is probably less convenient than
> using your distribution's package manager, at least initially. But
> it's a tradeoff: in return we can fix issues and make them available
> to you in a matter of minutes.
>

It looks OK for development, not so great for production where you do not
want compilers etc. installed,

I would be happy to help you with this if you can provide more
> details. In particular, you can instruct build2 to use system-
> installed libraries for any of the dependencies. The installation
> instructions[2] mention this explicitly:
>
> "Note that by default the underlying database libraries (libsqlite3,
> libmysqlclient, and libpq) will be built from source packages. If you would
> prefer to use the system-installed versions, adjust the corresponding build
> commands as follows:"
>
> $ bpkg build libodb-sqlite ?sys:libsqlite3
> $ bpkg build libodb-pgsql  ?sys:libpq
> $ bpkg build libodb-mysql  ?sys:libmysqlclient
>

Yes, this was my issue - build2 pulled in and built against later versions
than the platforms are distributed with.

Now CMake is cross platform (fully supported in WIndows VS) what does
build2 give you over and above CMake or Cmake + Conan (or similiar).


Regards
Sean
From boris at codesynthesis.com  Mon Nov 11 10:42:42 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Mon Nov 11 10:44:01 2019
Subject: [odb-users] ETA for ODB 2.5.0 (RC or release) packages
In-Reply-To: <CAHCKNJgFYznpmRVehk6U2y=DL2oPp+qM5mFr=-UjBALPOgqWPQ@mail.gmail.com>
References: <CAHCKNJgV9HzsDPWByLzria1F3N6Us8xE7kSKaz-P69=pTdCXVw@mail.gmail.com>
	<boris.20191105155720@codesynthesis.com>
	<CAHCKNJgFYznpmRVehk6U2y=DL2oPp+qM5mFr=-UjBALPOgqWPQ@mail.gmail.com>
Message-ID: <boris.20191111172649@codesynthesis.com>

Sean Clarke <sean.clarke@sec-consulting.co.uk> writes:

> Sorry, I should have been clearer - I was referring to the debian packages
> available on your website.. I gave up on the distros distributed version
> long ago for the reasons you give.

We have no plans to provide those after 2.4.0: building version-
independent (i.e., statically-linked) distribution packages is
more of a brittle hack than a long-term solution. And providing
version-specific package is beyond our capacity.


> It looks OK for development, not so great for production where you do not
> want compilers etc. installed,

If you don't want to build from source in production (understandably),
then provided your development (or build) machine is the same as your
production, you can build there, install everything into a separate
directory (see also config.install.chroot) and then ship the binaries
(either as simple .tar.?z or even custom deb/rpm/etc).


> Yes, this was my issue - build2 pulled in and built against later versions
> than the platforms are distributed with.

I don't think there is a general solution to this "my development
platform does not match my deployment" problem. With build2, if you
use the above method and build everything from source (as opposed
to using system-installed dependencies), you can probably get away
with it as long as your libc/libstdc++ are the same or ABI-compatible
(which is normally the case).

From ruffian_hy at 126.com  Fri Nov 15 00:07:32 2019
From: ruffian_hy at 126.com (=?GBK?B?u8beyA==?=)
Date: Fri Nov 15 00:09:36 2019
Subject: [odb-users] How to use the lazy_weak_ptr?
Message-ID: <3c8a0ec3.39ff.16e6d759335.Coremail.ruffian_hy@126.com>



I currently have two odb model like below:
A.hpp
class A
{
...
private:

odb::lazy_shared_ptr<B> m_b;
...

};
#pragma db object(A) pointer(std::shared_ptr)
#pragma db member(m_b)


B.hpp
class B
{
...
private:
odb::lazy_shared_ptr<A> m_a;
...
};
#pragma db object(B) pointer(std::shared_ptr)
#pragma db member(m_a)


They obviously have circular reference, and I want to use lazy_weak_ptr to fix it by the codes below:
A.hpp
class A
{
...
private:
odb::lazy_shared_ptr<B> m_b;
...
};
#pragma db object(A) pointer(std::shared_ptr)
#pragma db member(m_b)


B.hpp
class B
{
...
private:
#pragma db inverse(m_b)

odb::lazy_weak_ptr<A> m_a;
...
};
#pragma db object(B) pointer(std::shared_ptr)
#pragma db member(m_a)
 
But I get the error of ?error: object data member ? m_a' specified in db pragma column is inverse?.


It seems like I use the lazy_weak_ptr in a wrong way, is there anyone can help me?
From boris at codesynthesis.com  Fri Nov 15 13:52:50 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Fri Nov 15 13:54:24 2019
Subject: [odb-users] How to use the lazy_weak_ptr?
In-Reply-To: <3c8a0ec3.39ff.16e6d759335.Coremail.ruffian_hy@126.com>
References: <3c8a0ec3.39ff.16e6d759335.Coremail.ruffian_hy@126.com>
Message-ID: <boris.20191115204843@codesynthesis.com>

?? <ruffian_hy@126.com> writes:

> class B
> {
> ...
> private:
> #pragma db inverse(m_b)
> 
> odb::lazy_weak_ptr<A> m_a;
> ...
> };
> #pragma db object(B) pointer(std::shared_ptr)
> #pragma db member(m_a)
>  
> But I get the error of ?error: object data member ? m_a' specified in db pragma column is inverse?.

Try to move inverse to the member(m_a) pragma:

#pragma db member(m_a) inverse(m_b)

Also, dealing with circular relationship is tricky, especially when
spread over multiple headers; see "Circular Relationships" in the
manual for details:

https://codesynthesis.com/products/odb/doc/manual.xhtml#6.3

From ruffian_hy at 126.com  Fri Nov 15 20:31:56 2019
From: ruffian_hy at 126.com (=?GBK?B?u8beyA==?=)
Date: Fri Nov 15 20:33:43 2019
Subject: [odb-users] How to use the lazy_weak_ptr?
In-Reply-To: <boris.20191115204843@codesynthesis.com>
References: <3c8a0ec3.39ff.16e6d759335.Coremail.ruffian_hy@126.com>
	<boris.20191115204843@codesynthesis.com>
Message-ID: <3351360e.6a2.16e71d68b9d.Coremail.ruffian_hy@126.com>


I add #inlcude "B" and  move inverse to the member(m_a) pragma, but I still get the same error of ?error: object data member ? m_a' specified in db pragma column is inverse?.

At 2019-11-16 02:52:50, "Boris Kolpackov" <boris@codesynthesis.com> wrote:
>?? <ruffian_hy@126.com> writes:
>
>> class B
>> {
>> ...
>> private:
>> #pragma db inverse(m_b)
>> 
>> odb::lazy_weak_ptr<A> m_a;
>> ...
>> };
>> #pragma db object(B) pointer(std::shared_ptr)
>> #pragma db member(m_a)
>>  
>> But I get the error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
>
>Try to move inverse to the member(m_a) pragma:
>
>#pragma db member(m_a) inverse(m_b)
>
>Also, dealing with circular relationship is tricky, especially when
>spread over multiple headers; see "Circular Relationships" in the
>manual for details:
>
>https://codesynthesis.com/products/odb/doc/manual.xhtml#6.3
From ruffian_hy at 126.com  Fri Nov 15 20:51:34 2019
From: ruffian_hy at 126.com (=?GBK?B?u8beyA==?=)
Date: Fri Nov 15 20:53:10 2019
Subject: [odb-users] How to use the lazy_weak_ptr?
In-Reply-To: <boris.20191115204843@codesynthesis.com>
References: <3c8a0ec3.39ff.16e6d759335.Coremail.ruffian_hy@126.com>
	<boris.20191115204843@codesynthesis.com>
Message-ID: <6210e617.81f.16e71e884f5.Coremail.ruffian_hy@126.com>


I add #inlcude "B" and  move inverse to the member(m_a) pragma, but I still get the same error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
BTW, m_a also belong to a view join by this two class

class C { public: ... int64_t a; ... }; #pragma db view(C) pointer(std::shared_ptr) object(B) object(A: B::m_a == A::m_id) #pragma db member(C::a) column(B::m_a) 



At 2019-11-16 02:52:50, "Boris Kolpackov" <boris@codesynthesis.com> wrote:
>?? <ruffian_hy@126.com> writes:
>
>> class B
>> {
>> ...
>> private:
>> #pragma db inverse(m_b)
>> 
>> odb::lazy_weak_ptr<A> m_a;
>> ...
>> };
>> #pragma db object(B) pointer(std::shared_ptr)
>> #pragma db member(m_a)
>>  
>> But I get the error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
>
>Try to move inverse to the member(m_a) pragma:
>
>#pragma db member(m_a) inverse(m_b)
>
>Also, dealing with circular relationship is tricky, especially when
>spread over multiple headers; see "Circular Relationships" in the
>manual for details:
>
>https://codesynthesis.com/products/odb/doc/manual.xhtml#6.3





 
From ruffian_hy at 126.com  Fri Nov 15 21:22:36 2019
From: ruffian_hy at 126.com (=?GBK?B?u8beyA==?=)
Date: Fri Nov 15 21:24:08 2019
Subject: [odb-users] How to use the lazy_weak_ptr?
In-Reply-To: <boris.20191115204843@codesynthesis.com>
References: <3c8a0ec3.39ff.16e6d759335.Coremail.ruffian_hy@126.com>
	<boris.20191115204843@codesynthesis.com>
Message-ID: <710eacab.a49.16e7204ee60.Coremail.ruffian_hy@126.com>


I add #inlcude "B" and  move inverse to the member(m_a) pragma, but I still get the same error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
BTW, m_a also belong to a view join by this two class

class C { public: ... int64_t a; ... }; #pragma db view(C) pointer(std::shared_ptr) object(B) object(A: B::m_a == A::m_id) #pragma db member(C::a) column(B::m_a) 



At 2019-11-16 02:52:50, "Boris Kolpackov" <boris@codesynthesis.com> wrote:
>?? <ruffian_hy@126.com> writes:
>
>> class B
>> {
>> ...
>> private:
>> #pragma db inverse(m_b)
>> 
>> odb::lazy_weak_ptr<A> m_a;
>> ...
>> };
>> #pragma db object(B) pointer(std::shared_ptr)
>> #pragma db member(m_a)
>>  
>> But I get the error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
>
>Try to move inverse to the member(m_a) pragma:
>
>#pragma db member(m_a) inverse(m_b)
>
>Also, dealing with circular relationship is tricky, especially when
>spread over multiple headers; see "Circular Relationships" in the
>manual for details:
>
>https://codesynthesis.com/products/odb/doc/manual.xhtml#6.3





 





 
From ruffian_hy at 126.com  Fri Nov 15 21:27:29 2019
From: ruffian_hy at 126.com (=?GBK?B?u8beyA==?=)
Date: Fri Nov 15 21:29:03 2019
Subject: [odb-users] How to use the lazy_weak_ptr?
In-Reply-To: <710eacab.a49.16e7204ee60.Coremail.ruffian_hy@126.com>
References: <3c8a0ec3.39ff.16e6d759335.Coremail.ruffian_hy@126.com>
	<boris.20191115204843@codesynthesis.com>
	<710eacab.a49.16e7204ee60.Coremail.ruffian_hy@126.com>
Message-ID: <43f218e9.aa5.16e72096744.Coremail.ruffian_hy@126.com>



I move #pragma db member(C::a) column(B::m_a)  to "#pragma db member(C::a) column(A::m_id) "
But I get this below in the update schema:


        case 2:
        {
          db.execute ("UPDATE \"B\"\n"
                      "  SET \"a\" = NULL");
          db.execute ("UPDATE \"schema_version\"\n"
                      "  SET \"migration\" = 0\n"
                      "  WHERE \"name\" = ''");
          return false;
It will lost the existed data in db, right?



At 2019-11-16 10:22:36, "??" <ruffian_hy@126.com> wrote:


I add #inlcude "B" and  move inverse to the member(m_a) pragma, but I still get the same error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
BTW, m_a also belong to a view join by this two class

class C { public: ... int64_t a; ... }; #pragma db view(C) pointer(std::shared_ptr) object(B) object(A: B::m_a == A::m_id) #pragma db member(C::a) column(B::m_a) 



At 2019-11-16 02:52:50, "Boris Kolpackov" <boris@codesynthesis.com> wrote:
>?? <ruffian_hy@126.com> writes:
>
>> class B
>> {
>> ...
>> private:
>> #pragma db inverse(m_b)
>> 
>> odb::lazy_weak_ptr<A> m_a;
>> ...
>> };
>> #pragma db object(B) pointer(std::shared_ptr)
>> #pragma db member(m_a)
>>  
>> But I get the error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
>
>Try to move inverse to the member(m_a) pragma:
>
>#pragma db member(m_a) inverse(m_b)
>
>Also, dealing with circular relationship is tricky, especially when
>spread over multiple headers; see "Circular Relationships" in the
>manual for details:
>
>https://codesynthesis.com/products/odb/doc/manual.xhtml#6.3





 





 





 
From ruffian_hy at 126.com  Fri Nov 15 21:34:24 2019
From: ruffian_hy at 126.com (=?GBK?B?u8beyA==?=)
Date: Fri Nov 15 21:35:56 2019
Subject: [odb-users] How to use the lazy_weak_ptr?
In-Reply-To: <710eacab.a49.16e7204ee60.Coremail.ruffian_hy@126.com>
References: <3c8a0ec3.39ff.16e6d759335.Coremail.ruffian_hy@126.com>
	<boris.20191115204843@codesynthesis.com>
	<710eacab.a49.16e7204ee60.Coremail.ruffian_hy@126.com>
Message-ID: <396ad71a.b1f.16e720fbbb6.Coremail.ruffian_hy@126.com>






At 2019-11-16 10:22:36, "??" <ruffian_hy@126.com> wrote:


I add #inlcude "B" and  move inverse to the member(m_a) pragma, but I still get the same error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
BTW, m_a also belong to a view join by this two class

class C { public: ... int64_t a; ... }; #pragma db view(C) pointer(std::shared_ptr) object(B) object(A: B::m_a == A::m_id) #pragma db member(C::a) column(B::m_a) 



At 2019-11-16 02:52:50, "Boris Kolpackov" <boris@codesynthesis.com> wrote:
>?? <ruffian_hy@126.com> writes:
>
>> class B
>> {
>> ...
>> private:
>> #pragma db inverse(m_b)
>> 
>> odb::lazy_weak_ptr<A> m_a;
>> ...
>> };
>> #pragma db object(B) pointer(std::shared_ptr)
>> #pragma db member(m_a)
>>  
>> But I get the error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
>
>Try to move inverse to the member(m_a) pragma:
>
>#pragma db member(m_a) inverse(m_b)
>
>Also, dealing with circular relationship is tricky, especially when
>spread over multiple headers; see "Circular Relationships" in the
>manual for details:
>
>https://codesynthesis.com/products/odb/doc/manual.xhtml#6.3





 





 





 





 
From ruffian_hy at 126.com  Sat Nov 16 00:30:08 2019
From: ruffian_hy at 126.com (=?GBK?B?u8beyA==?=)
Date: Sat Nov 16 00:31:41 2019
Subject: [odb-users] How to use the lazy_weak_ptr?
In-Reply-To: <boris.20191115204843@codesynthesis.com>
References: <3c8a0ec3.39ff.16e6d759335.Coremail.ruffian_hy@126.com>
	<boris.20191115204843@codesynthesis.com>
Message-ID: <18408b3.14ad.16e72b0a075.Coremail.ruffian_hy@126.com>

I found that I can not use inverse in our project, 
But if I remove  ?#pragma db inverse(m_b)?, all the function below will get a nullptr
if(m_a.loaded())
auto post = m_a.get_eager().lock();
        auto post1=m_a.get_eager();
        auto post2 = m_a.load();
        auto post3 = m_a.lock();
        auto post4 = post3.load();
        auto post5 = post3.get_eager();
}


How can I get the object from lazy_weak_ptr?


At 2019-11-16 02:52:50, "Boris Kolpackov" <boris@codesynthesis.com> wrote:
>?? <ruffian_hy@126.com> writes:
>
>> class B
>> {
>> ...
>> private:
>> #pragma db inverse(m_b)
>> 
>> odb::lazy_weak_ptr<A> m_a;
>> ...
>> };
>> #pragma db object(B) pointer(std::shared_ptr)
>> #pragma db member(m_a)
>>  
>> But I get the error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
>
>Try to move inverse to the member(m_a) pragma:
>
>#pragma db member(m_a) inverse(m_b)
>
>Also, dealing with circular relationship is tricky, especially when
>spread over multiple headers; see "Circular Relationships" in the
>manual for details:
>
>https://codesynthesis.com/products/odb/doc/manual.xhtml#6.3
From ruffian_hy at 126.com  Sat Nov 16 00:49:48 2019
From: ruffian_hy at 126.com (=?GBK?B?u8beyA==?=)
Date: Sat Nov 16 00:51:21 2019
Subject: [odb-users] How to use the lazy_weak_ptr?
In-Reply-To: <18408b3.14ad.16e72b0a075.Coremail.ruffian_hy@126.com>
References: <3c8a0ec3.39ff.16e6d759335.Coremail.ruffian_hy@126.com>
	<boris.20191115204843@codesynthesis.com>
	<18408b3.14ad.16e72b0a075.Coremail.ruffian_hy@126.com>
Message-ID: <57a8ae47.15c5.16e72c2a08a.Coremail.ruffian_hy@126.com>

Actually, we have class Group and class Post, Group and Post is one to many relationship.
But we only need a most recent Post ptr for Group object.
Currently, we have odb::lazy_shared_ptr for each object, like below:
class Group
{
odb::lazy_shared_ptr<Post> mostRecentPost;
};
class Post
{
odb::lazy_shared_ptr<Group> group;
}
How can I break the ownership cycles for this scenario,  and prevent db upgrade incompatibility issues.



At 2019-11-16 13:30:08, "??" <ruffian_hy@126.com> wrote:

I found that I can not use inverse in our project, 
But if I remove  ?#pragma db inverse(m_b)?, all the function below will get a nullptr
if(m_a.loaded())
auto post = m_a.get_eager().lock();
        auto post1=m_a.get_eager();
        auto post2 = m_a.load();
        auto post3 = m_a.lock();
        auto post4 = post3.load();
        auto post5 = post3.get_eager();
}


How can I get the object from lazy_weak_ptr?


At 2019-11-16 02:52:50, "Boris Kolpackov" <boris@codesynthesis.com> wrote:
>?? <ruffian_hy@126.com> writes:
>
>> class B
>> {
>> ...
>> private:
>> #pragma db inverse(m_b)
>> 
>> odb::lazy_weak_ptr<A> m_a;
>> ...
>> };
>> #pragma db object(B) pointer(std::shared_ptr)
>> #pragma db member(m_a)
>>  
>> But I get the error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
>
>Try to move inverse to the member(m_a) pragma:
>
>#pragma db member(m_a) inverse(m_b)
>
>Also, dealing with circular relationship is tricky, especially when
>spread over multiple headers; see "Circular Relationships" in the
>manual for details:
>
>https://codesynthesis.com/products/odb/doc/manual.xhtml#6.3





 
From ruffian_hy at 126.com  Sat Nov 16 01:05:18 2019
From: ruffian_hy at 126.com (=?GBK?B?u8beyA==?=)
Date: Sat Nov 16 01:06:57 2019
Subject: [odb-users] How to use the lazy_weak_ptr?
In-Reply-To: <18408b3.14ad.16e72b0a075.Coremail.ruffian_hy@126.com>
References: <3c8a0ec3.39ff.16e6d759335.Coremail.ruffian_hy@126.com>
	<boris.20191115204843@codesynthesis.com>
	<18408b3.14ad.16e72b0a075.Coremail.ruffian_hy@126.com>
Message-ID: <6262e420.1712.16e72d0d38c.Coremail.ruffian_hy@126.com>

Actually, we have class Group and class Post, Group and Post is one to many relationship.
But we only need a most recent Post ptr for Group object.
Currently, we have odb::lazy_shared_ptr for each object, like below:
class Group
{
odb::lazy_shared_ptr<Post> mostRecentPost;
};
class Post
{
odb::lazy_shared_ptr<Group> group;
}
How can I break the ownership cycles for this scenario,  and prevent db upgrade incompatibility issues.



At 2019-11-16 13:30:08, "??" <ruffian_hy@126.com> wrote:

I found that I can not use inverse in our project, 
But if I remove  ?#pragma db inverse(m_b)?, all the function below will get a nullptr
if(m_a.loaded())
auto post = m_a.get_eager().lock();
        //auto post1=m_a.get_eager();
        auto post2 = m_a.load();
        auto post3 = m_a.lock();
        auto post4 = post3.load();
        auto post5 = post3.get_eager();
}


How can I get the object from lazy_weak_ptr?


At 2019-11-16 02:52:50, "Boris Kolpackov" <boris@codesynthesis.com> wrote:
>?? <ruffian_hy@126.com> writes:
>
>> class B
>> {
>> ...
>> private:
>> #pragma db inverse(m_b)
>> 
>> odb::lazy_weak_ptr<A> m_a;
>> ...
>> };
>> #pragma db object(B) pointer(std::shared_ptr)
>> #pragma db member(m_a)
>>  
>> But I get the error of ?error: object data member ? m_a' specified in db pragma column is inverse?.
>
>Try to move inverse to the member(m_a) pragma:
>
>#pragma db member(m_a) inverse(m_b)
>
>Also, dealing with circular relationship is tricky, especially when
>spread over multiple headers; see "Circular Relationships" in the
>manual for details:
>
>https://codesynthesis.com/products/odb/doc/manual.xhtml#6.3





 





 
From chere.loque at gmail.com  Fri Nov 22 09:10:43 2019
From: chere.loque at gmail.com (Vincent)
Date: Fri Nov 22 09:31:05 2019
Subject: [odb-users] Dependency issue with build2 0.12 and odb 2.5
Message-ID: <b0051360-2e31-5feb-84c8-c2c806ca24a3@gmail.com>

Hello!

I am facing a dependency issue when building odb beta 2.5 with build2
0.12.0. Everything worked correctly with build2 0.11.0, until I upgraded
to build2 0.12.0, which was required since I started to get this error 2
days ago:

   [...]
   + bpkg-stage --fetch-timeout 600 build --for install --yes --plan=
build2 bpkg bdep
   error: unable to satisfy constraint (build2 >= 0.12.0) for package build2
     info: available build2 version is 0.11.0
   info: while satisfying build2/0.12.0

Installed build2 like this:

   $ build2.sh --uninstall --yes
   $ curl -o build2.sh -sSfO
https://download.build2.org/0.12.0/build2-install-0.12.0.sh
   $ shasum -a 256 -b build2.sh
   $ sh build2.sh --no-check --trust yes --yes

   [...]
   About to download, build, and install build2 toolchain 0.12.0 (public).

   From:  https://download.build2.org
   Using: g++-8
   [...]
   Successfully installed build2 toolchain 0.12.0 (public).

Then, tried to install odb beta (2.5.0-b.17) :

   $ bpkg create -d odb-build2/ cc config.cxx=g++ config.cc.coptions=-O3
config.bin.rpath=my/local/lib/ config.install.root=my/local --wipe
   $ cd odb-build2/
   $ bpkg build --yes --trust-yes odb@https://pkg.cppget.org/1/beta
   fetching pkg:cppget.org/beta
   fetching pkg:cppget.org/testing (complements pkg:cppget.org/beta)
   fetching pkg:cppget.org/stable (complements pkg:cppget.org/testing)
   error: unable to satisfy constraint (build2 >= 0.11.0) for package odb
      info: available build2 version is 0.10.0
   info: while satisfying odb/2.5.0-b.17

Also tried with "/1/alpha" without success.

Thank you for your assistance!

Vincent

From boris at codesynthesis.com  Fri Nov 22 09:37:45 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Fri Nov 22 09:39:39 2019
Subject: [odb-users] Dependency issue with build2 0.12 and odb 2.5
In-Reply-To: <b0051360-2e31-5feb-84c8-c2c806ca24a3@gmail.com>
References: <b0051360-2e31-5feb-84c8-c2c806ca24a3@gmail.com>
Message-ID: <boris.20191122163256@codesynthesis.com>

Vincent <chere.loque@gmail.com> writes:

>    $ bpkg create -d odb-build2/ cc config.cxx=g++ config.cc.coptions=-O3
> config.bin.rpath=my/local/lib/ config.install.root=my/local --wipe
>    $ cd odb-build2/
>    $ bpkg build --yes --trust-yes odb@https://pkg.cppget.org/1/beta
>    fetching pkg:cppget.org/beta
>    fetching pkg:cppget.org/testing (complements pkg:cppget.org/beta)
>    fetching pkg:cppget.org/stable (complements pkg:cppget.org/testing)
>    error: unable to satisfy constraint (build2 >= 0.11.0) for package odb
>       info: available build2 version is 0.10.0
>    info: while satisfying odb/2.5.0-b.17

Hm, interesting. What do these commands print in the same terminal:

$ where bpkg
$ bpkg --version

$ where b
$ b --version

BTW, thanks for providing the detailed transcript, makes figuring
this out much easier.

From chere.loque at gmail.com  Fri Nov 22 11:00:05 2019
From: chere.loque at gmail.com (Vincent)
Date: Fri Nov 22 11:30:35 2019
Subject: [odb-users] Dependency issue with build2 0.12 and odb 2.5
In-Reply-To: <boris.20191122163256@codesynthesis.com>
References: <b0051360-2e31-5feb-84c8-c2c806ca24a3@gmail.com>
	<boris.20191122163256@codesynthesis.com>
Message-ID: <4fb546f8-1ad3-2902-d04f-cc3b22c8e8ab@gmail.com>

Thanks to your instructions, I managed to find out what happened!

We build/install build2 and odb libs in a subdirectory of our Jenkins
workspace, and all the calls to 'bpkg' from the makefiles use the
absolute path (I have not mentioned in my transcript for the sake of
clarity).

However, I found out that we also had 'bpkg' and 'b' binaries version
0.10 installed in the /usr/local/bin directory of the build machine.
Removing them solved the issue.

So, it seems that calling 'bpkg' in a specific path stills make
uses/calls of binaries installed at the system level (on the PATH), somehow.

Here is a full transcript of a test I made to reproduce the issue:

   $ cd /home/vincent/my_project

   $ cat ./build/dev-tools/build2/install_build2.sh
   SCRIPT_DIR=$( cd "$(dirname "${BASH_SOURCE}")" ; pwd -P )
   sh ${SCRIPT_DIR}/build2.sh --uninstall --yes
   curl -o ${SCRIPT_DIR}/build2.sh -sSfO
https://download.build2.org/0.12.0/build2-install-0.12.0.sh
   shasum -a 256 -b ${SCRIPT_DIR}/build2.sh
   sh ${SCRIPT_DIR}/build2.sh --no-check --trust yes --yes ${SCRIPT_DIR}

   $ ./build/dev-tools/build2/install_build2.sh

   About to uninstall build2 toolchain 0.11.0 (public).

   Install directory: /home/vincent/my_project/build/dev-tools/build2/
   Build directory:   /home/vincent/my_project/
   [...]
   Successfully uninstalled build2 toolchain 0.11.0 (public).

   da35b527aac3427b449ca7525e97f81faba776ae7680179521963b90a21fba12
*/home/vincent/my_project/build/dev-tools/build2/build2.sh

   -------------------------------------------------------------------------

   About to download, build, and install build2 toolchain 0.12.0 (public).
   [...]
   Install directory: /home/vincent/my_project/build/dev-tools/build2/
   Build directory:   /home/vincent/my_project/

   Package repository: https://pkg.cppget.org/1/alpha
   [...]
   Successfully installed build2 toolchain 0.12.0 (public).

   Install directory:   /home/vincent/my_project/build/dev-tools/build2/
   Build configuration: build2-toolchain-0.12/

Testing version installed in build directory:

   $ ./build/dev-tools/build2/bin/bpkg --version
   bpkg 0.12.0
   libbpkg 0.12.0
   libbutl 0.12.0
   Copyright (c) 2014-2019 Code Synthesis Ltd
   This is free software released under the MIT license.

   $ ./build/dev-tools/build2/bin/b --version
   build2 0.12.0
   libbutl 0.12.0
   host x86_64-linux-gnu
   Copyright (c) 2014-2019 Code Synthesis Ltd
   This is free software released under the MIT license.

And in system directory (/usr/local/bin):

   $ bpkg --version
   bpkg 0.10.0
   libbpkg 0.10.0
   libbutl 0.10.0
   Copyright (c) 2014-2019 Code Synthesis Ltd
   This is free software released under the MIT license.

   $ b --version
   build2 0.10.0
   libbutl 0.10.0
   host x86_64-linux-gnu
   Copyright (c) 2014-2019 Code Synthesis Ltd
   This is free software released under the MIT license.

Then, trying the actual build of odb:

   $ cd /home/vincent/my_project
   $ mkdir _build
   $ cd _build
   $ mkdir local
   $ mkdir odb-build2
   $ cd odb-build2
   $ ../../build/dev-tools/build2/bin/bpkg create -d . cc config.cxx=g++
config.cc.coptions=-O3 config.bin.rpath=../local/lib/
config.install.root=../local --wipe
   created new configuration in /home/vincent/my_project/_build/odb-build2/
   $ ../../build/dev-tools/build2/bin/bpkg build --yes --trust-yes
odb@https://pkg.cppget.org/1/beta
   added pkg:cppget.org/beta
   fetching pkg:cppget.org/beta
   fetching pkg:cppget.org/testing (complements pkg:cppget.org/beta)
   fetching pkg:cppget.org/stable (complements pkg:cppget.org/testing)
   error: unable to satisfy constraint (build2 >= 0.11.0) for package odb
     info: available build2 version is 0.10.0
   info: while satisfying odb/2.5.0-b.17

Is there any way to specify a "home path" for build2 binaries?

Vincent

From boris at codesynthesis.com  Fri Nov 22 11:35:55 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Fri Nov 22 11:37:50 2019
Subject: [odb-users] Dependency issue with build2 0.12 and odb 2.5
In-Reply-To: <4fb546f8-1ad3-2902-d04f-cc3b22c8e8ab@gmail.com>
References: <b0051360-2e31-5feb-84c8-c2c806ca24a3@gmail.com>
	<boris.20191122163256@codesynthesis.com>
	<4fb546f8-1ad3-2902-d04f-cc3b22c8e8ab@gmail.com>
Message-ID: <boris.20191122183000@codesynthesis.com>

Vincent <chere.loque@gmail.com> writes:

> So, it seems that calling 'bpkg' in a specific path stills make
> uses/calls of binaries installed at the system level (on the PATH), somehow.
> 
> [...]
>
>    $ ../../build/dev-tools/build2/bin/bpkg create -d . cc config.cxx=g++
> config.cc.coptions=-O3 config.bin.rpath=../local/lib/
> config.install.root=../local --wipe
>    created new configuration in /home/vincent/my_project/_build/odb-build2/
>    $ ../../build/dev-tools/build2/bin/bpkg build --yes --trust-yes
> odb@https://pkg.cppget.org/1/beta
>    added pkg:cppget.org/beta
>    fetching pkg:cppget.org/beta
>    fetching pkg:cppget.org/testing (complements pkg:cppget.org/beta)
>    fetching pkg:cppget.org/stable (complements pkg:cppget.org/testing)
>    error: unable to satisfy constraint (build2 >= 0.11.0) for package odb
>      info: available build2 version is 0.10.0
>    info: while satisfying odb/2.5.0-b.17
> 
> Is there any way to specify a "home path" for build2 binaries?

You could do:

$ PATH="../../build/dev-tools/build2/bin:$PATH" bpkg ...

But there is also --build <path>:

  The build program to be used to build packages. This should be the path to
  the build2 b executable. You can also specify additional options that should
  be passed to the build program with --build-option.[1]

So you can do:

$ ../../build/dev-tools/build2/bin/bpkg --build ../../build/dev-tools/build2/bin/b ...

[1] https://build2.org/bpkg/doc/bpkg-common-options.xhtml

From chere.loque at gmail.com  Fri Nov 22 11:55:25 2019
From: chere.loque at gmail.com (Vincent)
Date: Fri Nov 22 12:02:40 2019
Subject: [odb-users] Dependency issue with build2 0.12 and odb 2.5
In-Reply-To: <boris.20191122183000@codesynthesis.com>
References: <b0051360-2e31-5feb-84c8-c2c806ca24a3@gmail.com>
	<boris.20191122163256@codesynthesis.com>
	<4fb546f8-1ad3-2902-d04f-cc3b22c8e8ab@gmail.com>
	<boris.20191122183000@codesynthesis.com>
Message-ID: <89a618f7-3434-cdd3-0ca7-59ae1b3dbc18@gmail.com>

> But there is also --build <path>:
> 
>   The build program to be used to build packages. This should be the path to
>   the build2 b executable. [...]
> 
> So you can do:
> 
> $ ../../build/dev-tools/build2/bin/bpkg --build ../../build/dev-tools/build2/bin/b ...

Thank you, didn't notice this!

From odb at a-cunningham.com  Fri Nov 22 12:02:09 2019
From: odb at a-cunningham.com (Andrew Cunningham)
Date: Fri Nov 22 12:04:25 2019
Subject: [odb-users] Switch to clang?
Message-ID: <CAGTJuHjQurbT+TZJeRhRRhSJknX87F5G9k4iLge-iWKjMrEVgQ@mail.gmail.com>

Hi Boris,
Have you thought about switching to clang for your C++
parser/code-generator? We already use clang for code generation from our
C++ headers and it's very easy to use. Using clang has lots of benefits for
Windows developers as it's more of a native tool. ODB code generation is
painfully slow on Windows.

Andrew
From boris at codesynthesis.com  Mon Nov 25 08:05:45 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Mon Nov 25 08:07:50 2019
Subject: [odb-users] Switch to clang?
In-Reply-To: <CAGTJuHjQurbT+TZJeRhRRhSJknX87F5G9k4iLge-iWKjMrEVgQ@mail.gmail.com>
References: <CAGTJuHjQurbT+TZJeRhRRhSJknX87F5G9k4iLge-iWKjMrEVgQ@mail.gmail.com>
Message-ID: <boris.20191125145516@codesynthesis.com>

Andrew Cunningham <odb@a-cunningham.com> writes:

> Have you thought about switching to clang for your C++
> parser/code-generator?

This question comes up from time to time. It will be at least a
substantial effort. But it's also not clear if Clang has all the
capabilities that we need. One, in particular, is the ability to
instantiate templates after parsing of the translation unit is
complete. Last time I checks (which was admittedly a while ago),
this was impossible in Clang.


> ODB code generation is painfully slow on Windows.

Do you have any evidence that switching to Clang will significantly
improve compilation speed? In my experience, these days GCC's compiles
about as fast as Clang.

I am also assuming that you are using an optimized build of ODB (and
another thing that might help is running ODB compilation jobs in
parallel).

From blob84 at gmail.com  Sun Nov 24 13:01:19 2019
From: blob84 at gmail.com (blob blob)
Date: Mon Nov 25 08:13:56 2019
Subject: [odb-users] relationship always in a consistent state
Message-ID: <CALYX=rd=cMneUk-y+bauZxM88TOA4ceHCcFi8Y1Qv8x3rsWRbQ@mail.gmail.com>

Hello, I have not understood how make a bidirectional relationship always
in a consistent state.
How can I set both pointers at the same time?

Thanks.
From boris at codesynthesis.com  Tue Nov 26 05:21:34 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Tue Nov 26 05:23:40 2019
Subject: [odb-users] relationship always in a consistent state
In-Reply-To: <CALYX=rd=cMneUk-y+bauZxM88TOA4ceHCcFi8Y1Qv8x3rsWRbQ@mail.gmail.com>
References: <CALYX=rd=cMneUk-y+bauZxM88TOA4ceHCcFi8Y1Qv8x3rsWRbQ@mail.gmail.com>
Message-ID: <boris.20191126121730@codesynthesis.com>

blob blob <blob84@gmail.com> writes:

> I have not understood how make a bidirectional relationship always
> in a consistent state. How can I set both pointers at the same time?

You need to be more specific about what you want to achieve: Are you
talking about setting the pointers in the C++ object model or about
setting the corresponding foreign keys in the database. Both? Something
else entirely?

From blob84 at gmail.com  Tue Nov 26 13:24:22 2019
From: blob84 at gmail.com (blob blob)
Date: Tue Nov 26 13:26:47 2019
Subject: [odb-users] relationship always in a consistent state
In-Reply-To: <boris.20191126121730@codesynthesis.com>
References: <CALYX=rd=cMneUk-y+bauZxM88TOA4ceHCcFi8Y1Qv8x3rsWRbQ@mail.gmail.com>
	<boris.20191126121730@codesynthesis.com>
Message-ID: <CALYX=rdZhJDqiRFCVtTdBkXU=ETkBS2OOT+jYaeTCbdGi+_3Cg@mail.gmail.com>

I mean setting the pointers in the C++ object model, I want to update both
pointers at the same time, like the 6.2 of the odb manual.
I need a more concrete explanation.

Thank you.

Il giorno mar 26 nov 2019 alle ore 11:21 Boris Kolpackov <
boris@codesynthesis.com> ha scritto:

> blob blob <blob84@gmail.com> writes:
>
> > I have not understood how make a bidirectional relationship always
> > in a consistent state. How can I set both pointers at the same time?
>
> You need to be more specific about what you want to achieve: Are you
> talking about setting the pointers in the C++ object model or about
> setting the corresponding foreign keys in the database. Both? Something
> else entirely?
>
From boris at codesynthesis.com  Wed Nov 27 08:27:51 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Wed Nov 27 08:30:02 2019
Subject: [odb-users] relationship always in a consistent state
In-Reply-To: <CALYX=rdZhJDqiRFCVtTdBkXU=ETkBS2OOT+jYaeTCbdGi+_3Cg@mail.gmail.com>
References: <CALYX=rd=cMneUk-y+bauZxM88TOA4ceHCcFi8Y1Qv8x3rsWRbQ@mail.gmail.com>
	<boris.20191126121730@codesynthesis.com>
	<CALYX=rdZhJDqiRFCVtTdBkXU=ETkBS2OOT+jYaeTCbdGi+_3Cg@mail.gmail.com>
Message-ID: <boris.20191127152451@codesynthesis.com>

blob blob <blob84@gmail.com> writes:

> I mean setting the pointers in the C++ object model, I want to update both
> pointers at the same time, like the 6.2 of the odb manual.

The best you can do is make the data members private and then
have a modifier in one class that sets both pointers (you will
also need to make this class a friend of the other class so that
it can access the private member).

From alain.mari at se.com  Thu Nov 28 08:22:07 2019
From: alain.mari at se.com (ALAIN MARI)
Date: Thu Nov 28 09:15:04 2019
Subject: [odb-users] odb compiler and Windows SDK headers
Message-ID: <DB7PR04MB55454397172175AF7ABA3F65F1470@DB7PR04MB5545.eurprd04.prod.outlook.com>

We currently try to port our Windows software to ODB.
But many of our classes have specific Windows types like VARIANT which are defined in
Windows SDK include. These types concern only methods parameters types, not fields types that we want to persist.
As odb compiler uses its own headers (mingw), is it safe to try to include Windows headers also ?

Regards.

Alain.
From boris at codesynthesis.com  Thu Nov 28 09:21:04 2019
From: boris at codesynthesis.com (Boris Kolpackov)
Date: Thu Nov 28 09:23:20 2019
Subject: [odb-users] odb compiler and Windows SDK headers
In-Reply-To: <DB7PR04MB55454397172175AF7ABA3F65F1470@DB7PR04MB5545.eurprd04.prod.outlook.com>
References: <DB7PR04MB55454397172175AF7ABA3F65F1470@DB7PR04MB5545.eurprd04.prod.outlook.com>
Message-ID: <boris.20191128161546@codesynthesis.com>

ALAIN MARI <alain.mari@se.com> writes:

> But many of our classes have specific Windows types like VARIANT which
> are defined in Windows SDK include. These types concern only methods
> parameters types, not fields types that we want to persist. As odb
> compiler uses its own headers (mingw), is it safe to try to include
> Windows headers also ?

I think this should work for the commonly used stuff. In particular,
MinGW has a fairly complete Win32 API headers but more recent and/or
esoteric stuff (e.g., COM) is anyone's guess.

If you do run into issues, you can work around them by conditionally
excluding parts of your headers during ODB compilation with the
ODB_COMPILER macro. For example:

class object
{
#ifndef ODB_COMPILER

  ... // Windows-specific stuff.
  
#endif

private:

  ... // Data member.
};

It's not pretty but reliable.