[openib-general] win related [was: Re: [PATCH 1/2] opensm: sigusr1: syslog() fixes]

[Hal Rosenstock]

Hi Yevgeny,

On Wed, 2007-01-10 at 02:47, Yevgeny Kliteynik wrote:
> Hi Sasha,
> 
> Sasha Khapyorsky wrote:
> > Hi Yevgeny,
> > 
> > On 12:05 Tue 09 Jan     , Yevgeny Kliteynik wrote:
> >> Hi Sasha.
> >> Thanks for spotting this issue - I missed it when I reviewed the patch.
> >> Here's how things look in windows:
> >>  * openlog() and closelog() are redefined to nothing in windows config header
> >>  * cl_log_event() implementation on windows does all the work - gets handler
> >>    to event log, logs the message, and closes the handler.
> >>    It does *not* uses syslog().
> >>
> >> To use syslog/sysopen/sysclose on windows, a user have to install some 
> >> package (e.g. "Interix") that provides various daemons (such as syslogd), 
> >> similar to Linux/Unix:
> >> 	http://msdn2.microsoft.com/en-us/library/ms811897.aspx#ucmgch10_topic12
> >> 	http://www.interix.com/
> > 
> > And this is not done for WinOpenSM now. Is this would be useful option? I
> > guess this could solve a lot of compatibility problems and simplify the
> > porting dramatically, right?
> 
> I'll address this question to the windows guys. But I seriously doubt that
> they would want to make OSM depend on some third-party product, whether it
> is commercial or free.
>  
> >> So replacing cl_log_event() with syslog() would definitely cause a problem.
> >>
> >> A (pretty ugly) solution can be writing a wrapper function to cl_log_event() 
> >> on windows and redefining syslog() to use this wrapper.
> > 
> > Like #define syslog() cl_log_event() ? Not so ugly (better than
> > combination of openlog()/closelog() and "unrelated" cl_log_event()),
> > and perfectly simple IMO.
> 
> "#define syslog() cl_log_event()" will not do the job, since these functions 
> have different definitios:
> 	syslog(int priority, char *message, ...);
> 	cl_log_event(char*, cl_log_type_t, char*, void*, uint32_t);
> so there should be wrapper just for windows that will prepare the message
> for cl_log_event() as a single string.

So just so I am clear on this:

Is the wrapper approach acceptable to you ?

-- Hal
 
> >> But the question is: why do we need this replacement in the first place?
> > 
> > Look at the original Ira's patch for details - there syslog() is
> > introduced and closelog() is removed from cl_event_log(). I think his
> > motivation is to keep syslog connected for log reopening events and
> > prevent noisy openlog()/closelog() calls in cl_log_event().
> 
> Agree.
> 
> --Yevgeny
>  
> >>> replacement itself is needed in order to restore complib's cl_log_event()
> >>> behavior used by various applications (like ibutils ones and others).
> >> What behavior are we talking about here?
> > 
> > OTOH closelog() is needed than openlog()/syslog()/closelog() is part
> > shared library (look "Submitting Syslog Messages" at 'info libc' ).
> > 
> > Sasha
> >