Editing XSLT (section)

===Performance===
Most early XSLT processors were interpreters.  More recently, code generation is increasingly common, using portable intermediate languages (such as [[Java bytecode]] or .NET [[Common Intermediate Language]]) as the target. However, even the interpretive products generally offer separate analysis and execution phases, allowing an optimized expression tree to be created in memory and reused to perform multiple transformations. This gives substantial performance benefits in online publishing applications, where the same transformation is applied many times per second to different source documents.<ref>[http://www-128.ibm.com/developerworks/xml/library/x-xslt2/ Saxon: Anatomy of an XSLT processor] – Article describing implementation &amp; optimization details of a popular XSLT processor.</ref> This separation is reflected in the design of XSLT processing APIs (such as [[Java API for XML Processing|JAXP]]).

Early XSLT processors had very few optimizations.  Stylesheet documents were read into [[Document Object Model]]s and the processor would act on them directly. [[XPath]] engines were also not optimized. Increasingly, however, XSLT processors use optimization techniques found in functional programming languages and database query languages, such as static rewriting of an expression tree (e.g., to move calculations out of loops), and lazy pipelined evaluation to reduce the [[memory footprint]] of intermediate results (and allow "early exit" when the processor can evaluate an expression such as <code>following-sibling::*[1]</code> without a complete evaluation of all subexpressions). Many processors also use tree representations that are significantly more efficient (in both space and time)<ref>{{Cite journal|title=Improving Pattern Matching Performance in XSLT|first1=John|last1=Lumley|first2=Michael|last2=Kay|date=June 2015|journal=XML London 2015|doi=10.14337/XMLLondon15.Lumley01|url=http://xmllondon.com/2015/presentations/lumley|pages=9–25|doi-broken-date=1 November 2024 |isbn=978-0-9926471-2-4|doi-access=free}}</ref> than general-purpose DOM implementations.

In June 2014, Debbie Lockett and [[Michael Howard Kay|Michael Kay]] introduced an open-source benchmarking framework for XSLT processors called XT-Speedo.<ref>{{Cite journal|title=Benchmarking XSLT Performance|first1=Michael|last1=Kay|first2=Debbie|last2=Lockett|date=June 2014|journal=XML London 2014|doi=10.14337/XMLLondon14.Kay01|url=http://xmllondon.com/2014/presentations/kay|pages=10–23|doi-broken-date=1 November 2024 |isbn=978-0-9926471-1-7|doi-access=free}}</ref>