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

From: Daniel Walker (daniel.j.walker_at_[hidden])
Date: 2006-04-23 19:03:17

On 4/23/06, Peter Dimov <pdimov_at_[hidden]> wrote:
> Daniel Walker wrote:
> > On 4/23/06, Thorsten Ottosen <thorsten.ottosen_at_[hidden]> wrote:
> >> Daniel Walker wrote:
> >>>
> >>> Why is what better? Maybe I wasn't clear. When an XML file includes
> >>> a schema and fails validation when loaded, you do get a run-time
> >>> exception. I was trying to say that's a good thing. An XML
> >>> validating parser is similar to a compiler for a strongly typed
> >>> language: it catches type errors (in addition to syntax errors)
> >>> immediately before you actually try to use the file.
> >>
> >> Ok, why is that exception better than the one I generate if I don't
> >> meet the tag I expect, of if the contents of a tag is not of the
> >> type I expect?
> >
> > Well, you don't have to write C++ code to check if the tag/content is
> > what you expect. You declare the acceptable content (tags,
> > attribute/values, branch structures/sub-trees, etc) for your config
> > file in a schema, and the parser determines whether or not the XML
> > file conforms to the format declared in the schema. It has
> > time-saving, organization, and code-reuse advantages among others.
> > Most importantly, it has correctness advantages because one bug you
> > don't have to worry about when you're traversing the tree is running
> > into a tag you don't expect. By the time the file is loaded, all the
> > tags, tree structure, etc. has been validated by the parser and is
> > guaranteed to be the format you expect.
>
> I wouldn't omit the checks in the code, even when using a validating parser.
> When the schema and the program logic disagree, the program logic "wins". Or
> crashes. Either way, the schema loses. :-)

Good point about crashes. It is still possible to write programs that
crash even when using a schema. There is no silver bullet. Whether or
not you want to double check the parser/scheme is the same sort of
decision as whether or not you define NDEBUG in a released/deployed
system. Sure, the asserts are still useful, but do you really need the
extra check? The answer probably depends on the specific circumstances
and is somewhat a matter of taste. I think the more common case is
that you forget to manually check a constraint on the XML in your
code, in which case you would be glad if you had a validating parser.

Daniel Walker