jar scheme and xsl:include uniqueness
If you extract stylesheets from
test.jar and run
compile-xslt-tests.xsl, it works without any warning:
$ jar xf test.jar $ java -jar SaxonHE10-2J/saxon-he-10.2.jar -it -xsl:io/xspec/xspec/impl/src/compiler/compile-xslt-tests.xsl <?xml version="1.0" encoding="UTF-8"?>good
However, if you run the same stylesheet directly with
jar: scheme, an "included more than once" warning happens followed by a "duplicate named function" error:
$ java -jar SaxonHE10-2J/saxon-he-10.2.jar -it -xsl:'jar:file:/tmp/test.jar!/io/xspec/xspec/impl/src/compiler/compile-xslt-tests.xsl' -u Warning at xsl:stylesheet on line 6 column 31 of compile-scenario.xsl: Stylesheet module compile/compile-scenario.xsl is included or imported more than once. This is permitted, but may lead to errors or unexpected behavior Error at xsl:function on line 8 column 48 of compile-scenario.xsl: XTSE0770 Duplicate named function (see line 8 of jar:file:/tmp/test.jar!/io/xspec/xspec/impl/src/compiler/base/compile/compile-scenario.xsl) Error at xsl:call-template on line 9 column 52 of main.xsl: XTSE0650 Cannot find a template named compile-scenario Errors were reported during stylesheet compilation
xsl:include/@href be literally unique when loaded with
I searched the documentation but couldn't figure it out, except that a change history says, "it is possible to hold stylesheet modules in a JAR file and resolve xs:include and xs:import references between files within the same JAR archive."
Updated by Michael Kay almost 2 years ago
The problem appears to be as follows: URI resolution (of the relative URI against the base URI) is actually done twice. First we do it within the Saxon code, to decide whether this is a reference to a stylesheet module that we already know about. If it's not, then we call the URIResolver to fetch the stylesheet - and because the URIResolver is supplied with the relative URI and base URI separately, it does it again.
The URI resolution process within the URIResolver includes code that recognises the JAR: scheme specially (these are not actually standard RFC URIs, and they need custom handling: this is done by using the java.net.URL class in place of java.net.URI). But the code that computes a document key to access the stylesheet module cache is doing a straightforward call on
new URI(baseURI).resolve(href), and this gives the answer defined by RFC2396, which is not useful (the URI is classified as an "opaque" URI, so the rules of RFC2396 essentially mean that the relative URI is returned unchanged). The effect is that two
xsl:include references using the same relative URI are considered to refer to the same stylesheet module when this is not actually the case.
We need to change the code used to compute a document key so it uses the same algorithm as the standard URI resolver
Please register to edit this issue