Re: Problems sending SMSes with 6210

[Chris Kemp]

On 2003.03.05 10:25 BORBELY Zoltan wrote:
> Hi,
> 
> On Wed, Mar 05, 2003 at 08:18:22AM +0100, Ladislav Michl wrote:
> > On Tue, Mar 04, 2003 at 04:25:05PM +0100, BORBELY Zoltan wrote:
> > > Maybe we should put it into device.c, but currently only the serial
> > > driver requires this workaround.
> > 
> > perhaps better to say we received bugreport concerning serial driver
> only.
> > this is probably caused by fact that others are not so well tested.
> 
> No. The reason behind this bug is the different semantics of the write
> system call.
> 
> > > The link layer definitly the wrong place for it.
> > 
> > according my current understanding device layer only encapsulates
> > different devices. am i right? and if so read manual page for write
> > (AFAIR write on MS-Windows has the same behaviour). fix was done only
> > for serial driver. so doing this one layer above device one seems clear
> > to me in this situation.
> 
> As I stated earlier we should move it into device.c. It will fix this
> bug for all devices (link layer calls device_write instead of
> serial_write).
> 
> > now imagine some time critical mbus type protocol. link driver author
> > can call read between partial writes to read echoed bytes which will
> > lead to better behaviour not mentioning situation when you have also
> > small read buffer... (i'd like to do it for cbus as soon as
> statemachine
> > will be fixed properly after 0.5)
> 
> We should redesign the whole device layer/link layer/statemachine after
> 0.5.0. The current layout cannot fix a lot of important problems. Maybe
> we should create a multi threaded engine where a whole I/O handled by
> a device (or io) thread.

Aggh, we had a multithreaded solution initally.  One of the main points of
the current statemachine was to avoid threads!

Chris