[odb-users] Question concerning cross compiling ODB

[Boris Kolpackov]

Hi Michael,

Michael Powell <mwpowellhtx at gmail.com> writes:

> We've got an app that is cross compiled for Linux ARM architecture. If
> I understand properly, ODB hooks into the GCC in order to facilitate
> the code generation from the #pragmas embedded in the object model.

Yes, this is correct. Note however, that the ODB compilation is a
separate step and it doesn't have to use the same GCC (or, more
generally, the same C++ compiler) as the one you use to build your
applications.


> We've got GCC 4.7.2 cross compilers for ARM, I use a toolchain from
> launchpad.net for "headless" (non-IDE) building, and the Sourcery
> CodeBench toolchain for IDE-enriched development.
> 
> Possible to use ODB for this purpose?

Yes, sure. 


> What sort of cross compile support is there out of the box if, say,
> we download the sources, configure, make, and build it? How to hook
> it in with GCC?

The easiest way would be to use one of the pre-built ODB compiler
binaries that include a private copy of GCC (happens to also be 4.7.2).
We have a very detailed guide on how to cross-compile and use ODB for
the Raspberry Pi target (which also happens to be ARM/Linux).


> May also have some observations about how to abstract transactions
> away from common ORM code as a "unit of work" design pattern, perhaps
> using C++ templates: i.e. template<class UnitOfWork> type thing, with
> common operations being begin, commit, rollback, type thing.

I am not sure I understand what you are asking here. In ODB you don't
need to have a reference to transaction in order to perform a database
operation. Which means you can write code that just calls, say, load(),
as long as you know that it will be called within a transaction. In other
words:

// Has to be called within a transaction.
//
void nested (odb::database& db)
{
  ...

  shared_ptr<object> o (db.load<object> (id));

  ...
}

transaction t (db.begin ());
nested (db);
t.commit ();

Boris