Re: eclipse regression
Hi, On Fri, 2006-03-03 at 11:29 +0100, Mark Wielaard wrote: I haven't figured out since when this stopped working, but it clearly is a regression. If you installed the feature with a working version of classpath (0.20) and then try to use it with a CVS version you will get messages like: !ENTRY org.eclipse.update.configurator 2006-03-03 11:19:10.787 !MESSAGE Unable to parse file /tmp/eclipse/plugins/org.eclipse.cdt.core_3.0.2/plugin.xml. So I guess it is something going wrong with the parsing of xml files. But I haven't found the point yet where eclipse does this, or what the actual error is that occurs. Christian and I debugged this some more and finally found out that the following patch introduced the regression: SAX micropatch http://lists.gnu.org/archive/html/classpath-patches/2006-01/msg00277.html The patch seems completely reasonable. If we get any Exception so we fall out of the parse loop then we report a fatal error, cleanup and reset. Seems obvious since we will indeed not be able to continue with regular processing of the file. But we didn't count on Eclipse... Strangely enough the eclipse PluginParser is a DefaultHandler, that doesn't override fatalError(), and if it has enough data it will throw a subclass of SAXException from its own startElement() handler called ParseCompleteException. It tries to catch this exception from SAXParser.parse() and then continue happily. You can find that code at: http://dev.eclipse.org/viewcvs/index.cgi/*checkout*/org.eclipse.update.configurator/src/org/eclipse/update/internal/configurator/PluginParser.java?rev=HEADcontent-type=text/plain But of course since it threw this exception we, imho correctly, generate a SAXParseException and call fatalError() on the handler which will then throw this since the PluginParser didn't override it. I cannot really see a good way to make eclipse happy. It clearly expects that if it throws this SAXException from startElement() that it will get this exception back as if nothing happened. But since it has just destroyed our parser so we should indeed report a SAXParserException to it (through the errorHandler). Does anyone have a clever idea how to solve this? Cheers, Mark signature.asc Description: This is a digitally signed message part
Re: eclipse regression
On Sat, 2006-03-04 at 16:26 +0100, Mark Wielaard wrote:
I cannot really see a good way to make eclipse happy. It clearly expects
that if it throws this SAXException from startElement() that it will get
this exception back as if nothing happened. But since it has just
destroyed our parser so we should indeed report a SAXParserException to
it (through the errorHandler).
Thought a bit more about it and the following seems appropriate:
2006-03-04 Mark Wielaard [EMAIL PROTECTED]
* gnu/xml/stream/SAXParser.java (parse(InputSource)): Ignore
exceptions thrown by handlers while cleaning up and rethrow original
exception.
I checked with Christian that it solves the problem. So I am going to
commit this later and create the release branch, unless someone
complains loudly soon. (We can always fix it in a more appropriate way
if this is really too ugly for the next release.)
Cheers,
Mark
Index: gnu/xml/stream/SAXParser.java
===
RCS file: /cvsroot/classpath/classpath/gnu/xml/stream/SAXParser.java,v
retrieving revision 1.19
diff -u -r1.19 SAXParser.java
--- gnu/xml/stream/SAXParser.java 3 Mar 2006 12:30:59 - 1.19
+++ gnu/xml/stream/SAXParser.java 4 Mar 2006 17:54:38 -
@@ -657,17 +657,26 @@
}
catch (Exception e)
{
-if (!startDocumentDone contentHandler != null)
- contentHandler.startDocument();
SAXParseException e2 = new SAXParseException(e.getMessage(), this);
e2.initCause(e);
-if (errorHandler != null)
- errorHandler.fatalError(e2);
-if (contentHandler != null)
- contentHandler.endDocument();
+try
+ {
+if (!startDocumentDone contentHandler != null)
+ contentHandler.startDocument();
+if (errorHandler != null)
+ errorHandler.fatalError(e2);
+if (contentHandler != null)
+ contentHandler.endDocument();
+ }
+catch (SAXException sex)
+ {
+// Ignored, we will rethrow the original exception.
+ }
reset();
if (opened)
in.close();
+if (e instanceof SAXException)
+ throw (SAXException) e;
if (e instanceof IOException)
throw (IOException) e;
else
signature.asc
Description: This is a digitally signed message part
Re: eclipse regression
Hi Christian, On Fri, 2006-03-03 at 01:49 +0100, Christian Thalinger wrote: I'm unable to install some (any?) plugins in eclipse with CVS head. I already talked to tashiro on irc and it works with 0.20 release. The problem is in two places: * Software Updates - Find and install says: Current configuration problems * Software Updates - Manage Configuration does not allow to open Eclipse SDK, it only shows this entry. I hope i made myself clear (it's rather late). Yep. I am seeing the same thing. I was trying out http://developer.classpath.org/mediation/ClasspathHackingWithEclipse from scratch and trying to install the CDT plugin and it fails as you describe above. I haven't figured out since when this stopped working, but it clearly is a regression. If you installed the feature with a working version of classpath (0.20) and then try to use it with a CVS version you will get messages like: !ENTRY org.eclipse.update.configurator 2006-03-03 11:19:10.787 !MESSAGE Unable to parse file /tmp/eclipse/plugins/org.eclipse.cdt.core_3.0.2/plugin.xml. So I guess it is something going wrong with the parsing of xml files. But I haven't found the point yet where eclipse does this, or what the actual error is that occurs. Cheers, Mark signature.asc Description: This is a digitally signed message part
