Re: [boost] [Boost Review] Property Tree Library

From: Marcin Kalicinski (kalita_at_[hidden])
Date: 2006-04-23 05:18:23

>> However, I'm also sure you are aware of the fact that without _any_
> parsers
>> the library is quite useless. It might still have some obscure uses, but
> at
>> least 90% of it is gone.
>
> I'm not aware of this.  Please explain why the "property container" is
> useless without a parser.  This is the critcal reason why I have issues
> with this library in its present form.  Why is "property container"
> different
> from other containers such as vector or map in this regard.

Comparison with vector or another std container is a bad idea. property_tree 
is not usable as a container alone. For that you should rather use 
multi_index, as Gennadiy said many times - it would be more flexible, and 
faster as well. multi_index does not have get/put etc. interface, but if you 
populated the tree manually, you anyway have another way of accessing 
(properly typed) data, so get/put is not so important.

Empty property tree is useless. To populate it you need either to manually 
insert data, or use a function which does that in an organized way, like one 
of the parsers. While manually inserting data is common (in my experience 
with the library), it is only done to later save it to a file. Without 
parsers this becomes impossible.

Briefly, the strong point of the library is that it can represent data 
coming from different sources in an uniform way. You propose to cut that 
sources off completely, at least for a period of time.

How about adding the following question that each reviewer should try to 
answer:

Which of the existing parsers (XML, JSON, INI, Registry, cmdline, INFO) are 
mature enough to be included with the library now? Which should be improved, 
and undergo another review later?

When enough answers are gathered, the review manager will be able to make an 
educated decision what to cut.

Best regards,
Marcin