Re: [Boost-users] iostreams and serial ports help

From: Eric MALENFANT (Eric.Malenfant_at_[hidden])
Date: 2008-01-21 10:09:51

> 	Rob Desbois, le 18 janvier 2008 08:02:
> 
> 	>
> 	> After trying to compile I realised that a SeekableDevice
> 	> provides read, write and seek functions, nothing else.
> 	> Am I seriously misunderstanding the point of
> boost::iostreams?

> On Jan 18, 2008 6:46 PM, Eric MALENFANT
> <Eric.Malenfant_at_[hidden]> wrote:
 
> 	A file_descriptor only implements the Device concept,
> it not an std::iostream.
> 	If you want an std::iostream, you wrap the Device in a
> boost::iostreams::stream which, as stated in its
> documentation: "performs i/o by delegating to a contained
> Device", and "derives from a specialization of
> std::basic_istream, std::basic_ostream or
> std::basic_iostream, depending on whether the underlying
> Device models Source, Sink or both."

> 	Rob Desbois, le 18 janvier 2008 13:58:
>
> Are you able to comment on whether I 'should' be doing this
> as regards the non-blocking and non-seekable nature of a serial port?

The "seekability" should not be a problem. You could define a BidirectionalDevice as Jonathan suggests in another message. If you only need read (or write), you could even just define a Source (or Sink).

The "blocking" side of things may be more trouble, depending on your needs.
Boost.Iostreams offers minimal support for non-blocking strings: boost::iostreams::char_traits::would_block() can be used to indicate that a read or write attempt was interrupted but, unless I missed something, nothing allows the caller to wait for the blocking condition to be removed (like select() allows for non-blocking sockets, for example), so the caller has to poll the stream. As suggested by Bob Warren, Boost.Asio may offer an interesting alternative.

Éric Malenfant
---------------------------------------------
Amateur programmers think there are 1000 bytes in a kilobyte; 
Real Programmers think there are 1024 meters in a kilometer.