StackOverflow2013

Note that there are some explanatory texts on larger screens.

plurals

PO
text
Body
copied!<p>I am one of the developers of XQSharp, so I have experience in this area. XQSharp actually began its life as an XPath implementation before we expanded it to support XQuery.</p> <p>Our initial implementation took us about 6 months, although this was not the only thing we were working on at the time.</p> <p>After this time we had an implementation that was feature complete. There were many areas in which this was not fully conformant, where the standard .NET methods did not behave quite the same as the specification required. Some examples of this are with converting values to and from strings, regular expressions, a lot of unicode stuff, problems with the .NET representations of XML (eg handling of xml:base) and so on.</p> <p>There were several areas that needed to be done to implement this:</p> <p><strong>Parsing</strong>: The parser itself was straightforward, and mostly generated from the EBNF in the spec. I would estimate that this initially represented a few weeks work.</p> <p><strong>Data Model</strong>: How the data is represented. In order to have a full XPath implementation there are a lot of new data types (like xs:gDay) that need to be implemented. In our case we have all our items derive from a base type and all our expressions would return enumerators of these. You also need to be able to identify whether the type of an item matches a particular XPath type. We supported static typing and schema-awareness from the outset, without these features this section probably becomes trivial, but you are still looking at several weeks worth of work.</p> <p><strong>Expressions/Abstract Syntax Tree</strong> This is the model of the expression itself. We used the XQuery Formal Semantics document to produce a mapping from the various XPath constructs (for example axes and predicates) to a simpler core grammer (which ends up with huge amounts of let, for if and typeswitch expressions!). In our initial implementation all these expressions had evaluate methods and so were the final representation of the expression. In our case the expressions all had type check methods too, but this can be skipped initially (The main purpose of these is for optimization). Creating all these expressions again took several weeks.</p> <p><strong>Functions</strong> As a previous commenter pointed out the function library for XPath is rather large. The entire XPath library took us several months to implement.</p> <p><strong>Static Analysis</strong> A small amount of static analysis is required. Variable references and function calls must be bound to the correct variables and functions. Most XPath implementations are stack based, and so a stack allocation phase is required to assign pointers (or indexes) to all the variables. This static analysis took a week or two. The Dragon Book should set you up very nicely to solve most of these problems.</p> <p>You are probably looking at another month's worth of work for all the extra bits of work that do not fall directly into these categories.</p> <p>After all this work we were left with a mostly functional implementation of XPath; but it was far to slow for real world use (maybe 100x slower than XPath 1 in .NET). So after this comes the fun work - Optimization.</p> <p>Bringing the engine up to 100% conformance and adding optimizations probably took another 12-18 man months (although we probably went a little overboard with optimization!), but by that point we had already made the transition to being an XQuery implementation.</p> <p>My advice would be to start by tackling a subset of XPath (maybe forward axes only and a very limited function library) and you might be able to knock together an implementation in a month or two, but a serious XPath2 implementation will be a big investment in time.</p> <p>Make sure that you use XPathNavigator for your node representation, as it has methods like SelectChildren which can take advantages of indexes in the underlying representations (for example XPathDocument).</p>

Querying!

Guidance

An individual column

Larger individual text columns get their own page to allow for proper reading.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload