jkesselm 00/12/12 11:09:30
Modified: java/src/org/apache/xalan/processor
CompilingStylesheetHandler.java
Log:
Update to new template enumeration. May be awaiting fixes in the Composition
code.
Revision Changes Path
1.21 +117 -75
xml-xalan/java/src/org/apache/xalan/processor/CompilingStylesheetHandler.java
Index: CompilingStylesheetHandler.java
===================================================================
RCS file:
/home/cvs/xml-xalan/java/src/org/apache/xalan/processor/CompilingStylesheetHandler.java,v
retrieving revision 1.20
retrieving revision 1.21
diff -u -r1.20 -r1.21
--- CompilingStylesheetHandler.java 2000/12/06 05:27:23 1.20
+++ CompilingStylesheetHandler.java 2000/12/12 19:09:29 1.21
@@ -131,21 +131,63 @@
public void endDocument ()
throws org.xml.sax.SAXException
{
- try
- {
+
+// try
+// {
+
+ // Perform normal end-of-parse operations -- in particular,
+ // the "composition" process which performs data gathering
+ // and optimization of the interpretable tree.
super.endDocument();
- Vector compiledTemplates=new Vector();
-
- Stylesheet current=getStylesheet();
if(isStylesheetParsingComplete())
{
- // Begin compiling. Loop modeled on StylesheetRoot.recompose()
- // calling recomposeTemplates().
+ Stylesheet current=getStylesheet();
+
+ // Record the Templates as we compile them; this list gets
+ // passed into the "bundling" code to copy their excutable code
+ // into the .xsb file.
+ // TODO: Can we just use the StylesheetRoot's list?
+ Vector compiledTemplates=new Vector();
+
+ // cache the typecast
StylesheetRoot root=(StylesheetRoot)getStylesheetRoot();
- // loop from recompose()
- int nImports = root.getGlobalImportCount();
+ // For all templates used in this stylesheet (including
+ // those from imported and included stylesheets),
+ // compile to Java. (Takes advantage of the fact that
+ // ComposedStylesheet processing has already gathered
+ // these and reconciled conflicts.)
+
+ // New TemplateList being generated
+ org.apache.xalan.templates.TemplateList newTl
+ = new org.apache.xalan.templates.TemplateList();
+ // Iterate over contents of old TemplateList
+ org.apache.xalan.templates.TemplateList.TemplateWalker tw
+ =root.getTemplateListComposed().getWalker();
+
+ // Scan all templates in old list, compile, insert into new
+ ElemTemplate et;
+ while ( null != (et = tw.next()) )
+ {
+ ElemTemplate ct = compileTemplate(et);
+
+ // If compilation succeeds, use it; else fall back on
interp
+ newTl.setTemplate( (ct!=null) ? ct : et);
+ }
+ // Postprocess/reconcile list
+ newTl.compose();
+ // And make it active
+ root.setTemplateListComposed(newTl);
+
+ // TODO: Theoretically, we can now discard imports/includes
+ // -- they're no longer pointing at the right ElemTemplates
+ // anyway. There's discussion of doing so in any case when
+ // they aren't needed for tooling support. Always do it here,
+ // leave it to be handled there, or other?
+
+/****** OLD APPROACH
+ int nImports = root.getGlobalImportCount();
for(int imp = 0; imp < nImports; imp++)
{
org.apache.xalan.templates.StylesheetComposed
@@ -185,17 +227,31 @@
// entry point... Or flush might be made the new default
// behavior; I don't know whether that would be appropriate.
root.recomposeTemplates(true);
-
- // TODO: Should bundling occur elsewhere?
+ ****** OLD APPROACH */
+
+ // TODO: Theoretically, we no longer need the Imported trees.
+ // Eliminating them would save some bytes in memory and in the
+ // .xsb bundle file. (They should be needed only for tooling
+ // support, which doesn't seem to apply to compiled operation.)
+ // Try that _AFTER_ the compile loop has been revived!
+
+ // TODO: Should bundling occur elsewhere or be optional?
CompiledStylesheetBundle.createBundle(root,compiledTemplates);
}
- }
- catch(TransformerException te)
- {
- throw new org.xml.sax.SAXException(te);
- }
+
+// }
+// catch(TransformerException te)
+// {
+// throw new org.xml.sax.SAXException(te);
+// }
}
+
+ // Minor optimization: Only need to look these up once
+ private static final org.apache.xml.utils.synthetic.Class
sClassCompiledTemplate=org.apache.xml.utils.synthetic.Class.forClass(org.apache.xalan.processor.CompiledTemplate.class);
+ private static final org.apache.xml.utils.synthetic.Class
sClassObjectArray=org.apache.xml.utils.synthetic.Class.forClass(Object[].class);
+
+
/**
Analyse an <xsl:template> tree, converting some (most?)
of its recursive evaluate() behavior into straight-line
@@ -220,15 +276,9 @@
to the attribute trees.
xsl:choose and xsl:for-each may be flattened
Namespace declarations become SAX prefix operations.
- ***** NS aliasing really should have occured before we get here,
- as should most NS resolution. The annoying exception is when
- xsl:attribute has a prefixed name but no explicit namespace,
- a feature which I Really Wish had not been supported.
-
+ <p>
Other nodes simply have their .evaluate() invoked
TODO: Their children really should be walked for further compilation
opportunities.
- TODO: ***** OPTIMIZATION: We should preload/cache the
org.apache.xml.utils.synthetic.Class
- objects rather than doing forName/forClass lookups every time.
*/
ElemTemplate compileTemplate(ElemTemplate source)
{
@@ -243,7 +293,7 @@
org.apache.xml.utils.synthetic.Class tClass=
org.apache.xml.utils.synthetic.Class.declareClass(className);
tClass.setModifiers(java.lang.reflect.Modifier.PUBLIC);
-
tClass.setSuperClass(tClass.forName("org.apache.xalan.processor.CompiledTemplate"));
+ tClass.setSuperClass(sClassCompiledTemplate);
// public constructor: Copy values from original
// template object, pick up "uncompiled children"
@@ -252,7 +302,7 @@
tClass.declareConstructor();
ctor.setModifiers(java.lang.reflect.Modifier.PUBLIC);
ctor.addParameter(tClass.forClass(ElemTemplate.class),"original");
-
ctor.addParameter(tClass.forName("java.lang.Object[]"),"interpretArray");
+ ctor.addParameter(sClassObjectArray,"interpretArray");
ctor.getBody().append(
"super(original,\n"
+'\t'+source.getLineNumber()+','+source.getColumnNumber()+",\n"
@@ -337,11 +387,11 @@
}
- // Compile the new java class
+ // Compile the new java class. Note that any existing class
+ // by the same name will be walked upon... which is why I'm
+ // going to the trouble of generating unique classnames.
// TODO: ***** ISSUE: Where write out the class? Parameterize.
- // Needs to be somewhere on the classpath.
- // TODO: ***** ISSUE: What if file already exists?
- // I think the answer in this case is "overwrite it.".
+ // Needs to be somewhere on the classpath, or we need a classloader.
Class realclass=compileSyntheticClass(tClass,".");
// Prepare the array of execute()ables
Object[] eteParms=new Object[interpretVector.size()];
@@ -358,11 +408,6 @@
System.out.println("CompilingStylesheetHandler class synthesis
error");
e.printStackTrace();
}
- catch(java.lang.ClassNotFoundException e)
- {
- System.out.println("CompilingStylesheetHandler class resolution
error");
- e.printStackTrace();
- }
catch(java.lang.IllegalAccessException e)
{
System.out.println("CompilingStylesheetHandler class comilation
error");
@@ -397,9 +442,10 @@
compileElemLiteralResult((ElemLiteralResult)kid,body,interpretVector);
break;
- // TODO: ***** Redirection of attr value not working yet.
- //case Constants.ELEMNAME_ATTRIBUTE:
- // compileElemAttribute((ElemAttribute)kid,body,interpretVector);
+ // TODO: ***** Redirection of attr value not working yet.
+ // TODO: ***** Attrs should be preprocessed (SAX-ordered)
+ //case Constants.ELEMNAME_ATTRIBUTE:
+ // compileElemAttribute((ElemAttribute)kid,body,interpretVector);
// break;
default:
@@ -460,7 +506,7 @@
}
// Process AVTs.
- // TODO: Should be be checking for excluded namespace prefixes?
+ // TODO: Should we be checking for excluded namespace prefixes?
// ***** That wasn't done in non-compiled version, but was an open issue.
java.util.Enumeration avts=ele.enumerateLiteralResultAttributes();
if (avts!=null)
@@ -486,8 +532,7 @@
else
{
// Expression. Must resolve at runtime.
- // TODO: ***** It might be possible to unwind this
- // evaluation too. Consider.
+ // TODO: ***** It might be possible to unwind this too.
int offset=interpretVector.size();
interpretVector.addElement(avt);
body.append(
@@ -606,11 +651,11 @@
// YES, this is inconsistant.
int offset=interpretVector.size();
interpretVector.addElement(avt);
- // TODO: ***** I'm assuming I can get away with "this" as the
prefixResolver
- // even in the compiled code. I'm not really convinced that's true...
return
- "(
((org.apache.xalan.templates.AVT)m_interpretArray["+offset+"]).evaluate(transformer.getXPathContext(),sourceNode,this,new
StringBuffer()) )"
- ;
+ "( ((org.apache.xalan.templates.AVT)m_interpretArray["
+ +offset
+
+"]).evaluate(transformer.getXPathContext(),sourceNode,this,new StringBuffer())
)"
+ ;
}
// Wrap quotes around a text string.
@@ -699,7 +744,7 @@
// If they are trying to add an attribute when there isn't an
// element pending, it is an error.
- // TODO: ****** Could these tests occur _before_ we transform?
+ // TODO: ****** Could these tests occur _before_ we transform? Factor?
body.append(
"if(null == rhandler.getPendingElementName())\n"
+"{\n"
@@ -834,13 +879,12 @@
kid!=null;
kid=kid.getNextSiblingElem())
{
- //TODO: *PROBLEM* NEED EQUIVALENT? Node is Going Away...
- // body.append("transformer.pushElemTemplateElement(kid);\n");
-
+ //TODO: NEED EQUIVALENT? This Node is Going Away...
+ // body.append("transformer.pushElemTemplateElement(kid);\n");
compileElemTemplateElement(kid,body,interpretVector);
- //TODO: *PROBLEM* NEED EQUIVALENT? Node is Going Away...
- // body.append("transformer.popElemTemplateElement(kid);\n");
+ //TODO: NEED EQUIVALENT? This Node is Going Away...
+ // body.append("transformer.popElemTemplateElement(kid);\n");
}
// End the class wrapper
@@ -862,7 +906,7 @@
* name and output is written to the directory that file
* would be found in (possibly relative). However, "."
* is treated as being found in itself rather than in "..".
- * TODO: ***** A more elegant version of this should be moved into
org.apache.xml.utils.synthetic.Class?
+ * TODO: ***** Move polished version into
org.apache.xml.utils.synthetic.Class
* TODO: Should we use a classloader rather than std. classpath?
*/
Class compileSyntheticClass(org.apache.xml.utils.synthetic.Class tClass,
String classLocation)
@@ -910,21 +954,21 @@
return null;
}
- // Try to pick up the same classpath we're executing under. That
- // ought to include everything in Xalan and the standard libraries.
+ // Try to pick up the same classpath we're executing under. That
+ // ought to include everything in Xalan and the standard libraries.
String classpath=System.getProperty ("java.class.path");
-
- // If compiling with the -g switch (Java debugging), we should
retain
- // the Java source code to support debugging into the synthesized
class.
- // Some additional diagnostics are also turned on as a side effect.
- // TODO: Find a better place to put the debugging control.
- String javac_options=
-
System.getProperty("org.apache.xalan.processor.CompilingStylesheetHandler.options","");
- boolean debug=(javac_options.indexOf("-g")>=0);
-
- // Run the compilation. Encapsulates the fallbacks and
- // workarounds needed to achieve this in various environments.
- JavaUtils.setDebug(debug);
+
+ // If compiling with the -g switch (Java debugging), we should retain
+ // the Java source code to support debugging into the synthesized class.
+ // Some additional diagnostics are also turned on as a side effect.
+ // TODO: Find a better place to put the debugging control. Property?
Parm?
+ String javac_options=
+
System.getProperty("org.apache.xalan.processor.CompilingStylesheetHandler.options","");
+ boolean debug=(javac_options.indexOf("-g")>=0);
+
+ // Run the compilation. Encapsulates the fallbacks and
+ // workarounds needed to achieve this in various environments.
+ JavaUtils.setDebug(debug);
boolean compileOK=
JavaUtils.JDKcompile(filename,classpath);
@@ -981,7 +1025,7 @@
//which need to be made globally unique if they aren't
//to collide with other compilations loaded into the same
//JVM. We really need classnames based on something like
- //UUID or GID, combining the source stylesheet's URI,
+ //UUID or GID, combining the source stylesheet's URI and locator,
//where it was processed, exactly when it was processed
//... and then going the extra step to guard against
//multitasking causing two stylesheets to start within
@@ -1007,22 +1051,20 @@
long templateNumber;
synchronized(this)
- { // I don't know if ++ is atomic or not. I'd hope so,
- // but until that's been checked we should synchronize
- // this operation.
+ { // Atomic get-next-count -- last-ditch protection against
+ // the risk of multiple threads trying to process the same
+ // stylesheet at the same instant on the same machine.
templateNumber = ++templateCounter;
}
// Generate a mostly-unique name.
// TODO: ***** NOT GUARANTEED TO BE COMPLETELY UNIQUE, since
- // there could, theoretically, be two seperate processes
- // which are running this code simultaneously -- so
- // even the combination of server, timestamp, and
- // unique index may not suffice. The officious answer
- // would be to tap into a system-level unique ID service,
+ // there could, theoretically, be two seperate PROCESSES
+ // which are running this code simultaneously. The officious
+ // answer would be to tap into a system-level unique ID service,
// like UUID or GID.
// ***** Unclear these names will be supportable on all
- // systems. Some may have limits on filename length.
+ // systems. Some platforms may have limits on filename length.
// Could move more of the fields into directory names if
// that helps.
String className=
