maria-developers team mailing list archive

[Frank Fegert <fra.nospam.nk@xxxxxx>]

Hello Wlad,

thanks for your fast reply!

On Sun, Jun 23, 2013 at 02:26:16PM +0200, Vladislav Vaintroub wrote:
> I think that the realistic options  you have are : 
> 1. Either you have an AIX enthusiast,  a knowledgeable  guy who regularly
> supplies team with AIX related patches (that do not break  any other
> platforms;)). 

An AIX enthusiast i am and more knowledge is always good, it's just those
48h days no crazy physicist wants to come up with ;-)

> Or 
> 2. You provide a machine for the Buildbot, and sponsor the project (the
> popularity of AIX is IMO too low, to support it for free). 
> 
> It is my personal opinion, not an official position.

Though not official, still very much appreciated! A machine would be no
problem at all, network access is a different story though, i'll have to
check with our security guys. My machines @home would be a last resort,
you'd hate me for my low bandwidth though ;-) Sponsoring could be a bit
complicated, since a) my lottery ticket didn't pan out - again ;-) and
b) we're in the public sector in germany, so lots of bureaucracy and our
procurement department is bound by some very "odd" rules. I'll talk to
my team leader and see what we can do.

> To me , it looks like a  sound alternative for AIX environment  would be
> using IO completion ports, the AIX port of  Windows API's.
> Native_event  would be OVERLAPPED *
> Io_port_create, io_port_associate_fd map to  CreateIoCompletionPort () 
> io_poll_start_read is  ReadFile(), with OVERLAPPED*, 0 bytes length read
> io_poll_wait is GetQueuedCompletionStatus()
> io_port_disassociate() cannot be implemented, but it is only used in rare
> case where thread_pool_size is being changed at runtime.
> 
> If you do not want to mess with it (you said your C-fu is weak),  I attached
> a patch  with dummy implementation of threadpool required methods, it should
> just make this thing compile, threadpool won't be usable. I did not test it.
> 
> PS. Performance schema unit tests should either be disabled or link with sql
> library (TARGET_LINK_LIBRARIES sql), I gather this from the error output you
> sent.

Thanks for the pointers, i'll see if i can come up with some IOCP code.
If not i'll try the fallback you provided. I'll report back as soon as
possible.

Thanks again & best regards,

    Frank