From: Peter Dimov (pdimov_at_[hidden])
Date: 2020-09-22 13:17:09
Bjorn Reese wrote:
> On 2020-09-22 14:20, Peter Dimov via Boost wrote:
>
> > I don't think so. In fact the reverse is true. Deserialization from DOM
> > is trivial. Making a deserialization library build a DOM is decidedly
> > nontrivial. I'm not sure how it could be done.
>
> It is easy with a pull parser. The following header shows both direct
> serialization from DOM to JSON, and direct deserizalization from JSON to
> DOM:
>
> https://github.com/breese/trial.protocol/blob/develop/include/trial/protocol/json/serialization/dynamic/variable.hpp
Yes, you're right, it's indeed easy with a pull parser, if the value and the
archive cooperate.
This still doesn't make one approach a superset of the other though. You can
feed Boost.JSON's push parser incrementally. A pull parser, being a pull
one, reverses the flow of control and asks you (its stream) for data when
you pull from it. Yes, it makes things like the above possible, but I don't
think it entirely supersedes pull parsers.
I remember this being brought up before, so you may have a solution for the
incremental case. A pull parser could f.ex. return token::need_input or
something like that when it's starved, like a nonblocking socket. There'll
still be a disconnect between its buffer size and the amount of incoming
data though, unless I'm missing something.