Hello,

Le 10/12/2009 17:36, Peter Soetens a écrit :
> That means a lot ! Also don't forget to run the list-test since it
> tests the lock-free lists.

The list-test passed also. However, I think this test only uses
ListLocked which is mutex based ?

Since you did not push anything on the main repository yet, I changed
some of my commit messages (since the remote changed deeply you might
have to clone again the branch to avoid any problems. I put the changes
in the "for_peter" branch to avoid any confusion):

> - could you prefix your commits in the future with a 'tag', like
> win32: Export missing symbols or cmake: Hide lockfree structures with
> OROBLD_OS_NO_ASM etc.

Done. I must admit sometimes I did not know which one to pick :)

> - When adding a multi-line commit message, be sure to leave a blank
> line between the first and third line, because git am and other tools
> interprete commit messages as email bodies. The subject and body in
> emails are separated by a blank line.

Done.

> - Is the simple Posix wrapper DLL present on all windows platforms ?
> What else does it include ?

Yes, it is the standard WinSock2/"Compatibility" interface.

> - /MP is not yet officially supported on MSVC 2005. It works 'kindof',
> sometimes you have to turn it off. Maybe turn it off by default if
> MSVC80 cmake variable is true ?

Ok, I put a MSVC condition on that.

> Finally, I'm planning to extend the 'plugin' functionality of the RTT
> with functions for loading DLL/.so files. This will be used to load
> tool/type kits and components by RTT of the deployer. This would make
> the deployment component easier to port to win32. Do you have any
> advise on how to do this best on win32 ? There was before a discussion
> on the mailinglist that there are dlopen wrappers for win32, but maybe
> you prefer a full-native solution ?

I'm not a specialist of DL on windows, but I think that the wrapping of
dlopen,dlclose,dlsym,... is pretty straightforward using LoadLibrary,
FreeLibrary,GetProcAddress,... Maybe a very simple wrapper like the one
is the os/* directories can be a simple way to do that. However, if you
want to have a kind of "universal binary module", for instance loading
.dll under linux or .so under windows, it's tricky but acheivable (cf.
Wine), but might not be worth the effort.

> I see no reasons to delay accepting your branch, but since I won't
> merge it on a daily basis, I'll wait for your final ok-to-pull, in
> case you would change something in response to my above remarks.

It OK for me !

We already use RTT-1.10 on Windows/MSVC 2005 and everything seems to work
well. Excuse my ignorance, but what does this patch brings more compared to
the current version?

Thank you,

Philippe

2009/12/10 Peter Soetens <peter [..] ...>

> Hi Jean,
>
> On Thu, Dec 10, 2009 at 16:19, Jean Sreng <jean [dot] sreng [..] ...> wrote:
> > Le 04/12/2009 09:29, Peter Soetens a écrit :
> >>> I would like to know if you are interested in this changeset and if so,
> >>> in which form (patches, fork on GitHub, etc.). Some of my code may need
> >>> some work to fit your coding standards although...
> >>
> >> If you're familiar with github, it's easiest for me if you fork my
> >> repository and create a for-peter branch which I can pull and merge.
> >> If you're not into git, a patch on the mailinglist will be just fine
> >> too.
> >
> > Hello, I just pushed on github (http://github.com/jsreng/orocos-rtt) the
> > atomic functions for MSVC. I didn't tested them heavily but core-test
> > passed without any errors.
>
> That means a lot ! Also don't forget to run the list-test since it
> tests the lock-free lists.
>
> I could merge your branch locally on the rtt-2.0 mainline. All seems
> fine, but these minor questions:
> - could you prefix your commits in the future with a 'tag', like
> win32: Export missing symbols or cmake: Hide lockfree structures with
> OROBLD_OS_NO_ASM etc.
> - When adding a multi-line commit message, be sure to leave a blank
> line between the first and third line, because git am and other tools
> interprete commit messages as email bodies. The subject and body in
> emails are separated by a blank line.
> - Is the simple Posix wrapper DLL present on all windows platforms ?
> What else does it include ?
> - /MP is not yet officially supported on MSVC 2005. It works 'kindof',
> sometimes you have to turn it off. Maybe turn it off by default if
> MSVC80 cmake variable is true ?
>
> Finally, I'm planning to extend the 'plugin' functionality of the RTT
> with functions for loading DLL/.so files. This will be used to load
> tool/type kits and components by RTT of the deployer. This would make
> the deployment component easier to port to win32. Do you have any
> advise on how to do this best on win32 ? There was before a discussion
> on the mailinglist that there are dlopen wrappers for win32, but maybe
> you prefer a full-native solution ?
>
> I see no reasons to delay accepting your branch, but since I won't
> merge it on a daily basis, I'll wait for your final ok-to-pull, in
> case you would change something in response to my above remarks.
>
> Thanks for doing this !
>
> Peter
> --
> Orocos-Dev mailing list
> Orocos-Dev [..] ...
> http://lists.mech.kuleuven.be/mailman/listinfo/orocos-dev
>

On Thu, Dec 10, 2009 at 17:48, Philippe Hamelin
<philippe [dot] hamelin [..] ...> wrote:
> We already use RTT-1.10 on Windows/MSVC 2005 and everything seems to work
> well. Excuse my ignorance, but what does this patch brings more compared to
> the current version?

If I may answer.
The current Windows target uses mutex objects to synchronize access to
thread-safe buffers, lists, queues and data objects. The patches of
Jean are 1) for the RTT 2.0 mainline and 2) allow to use the lock-free
algorithms instead of the mutexes. There are also some cleanups that
could be back-ported to the 1.x lines though.

Peter