Boost logo

Boost :

From: Reece Dunn (msclrhd_at_[hidden])
Date: 2003-12-17 10:34:19

There has been recent discussion about the similarity between serialization
and display. In response to this I have implemented an experimental data
input mechanism for text-based systems in my library:

If you want binary input/output or better input performance, you should use
a serialization library as has been discussed elsewhere.

I have several questions:

[1] Is there a demand for input support of complex/sequential data types
from a stream?

[2] What are peoples views on my implementation? Should I continue?

[3] Is there a need for better exception/error information from a failed
input operation -- incompatible data types; size of list not supplied; data
type to small to fit the data; etc.?

[4] When implementing an input operation on a stream, it is advised/required
you should not modify the object passed to it unless the operation succeeds.
My feeling on this is that it would lead to inefficiencies (i.e. excessive
copying). I want to construct the object on the fly, leaving a partially
constructed object on failure, to improve efficiency. I will strongly
document this deviated behaviour, but what do other people feel about this?

Documentation will be provided if there is interest in this addition.

Aside from this addiion, do people feel that my library is ready for review.
I intend to produce tutorials to compliment the documentation, but apart
from that, is there anything lacking?

Reece H Dunn

Tired of 56k? Get a FREE BT Broadband connection

Boost list run by bdawes at, gregod at, cpdaniel at, john at