The supplied InputStream in unnecessarily written into memory about validator HOT 3 CLOSED

Indeed this is supported by saxon and is definitely a good idea to reduce memory usage of the validator. Changings this requires some rewrite of Input and InputFactory but this should not break the api.

Nevertheless, I tested a litte bit and have trouble at least with the SchemaValidationAction which is not producing errors with line and col information. I reused the XdmNode parsed in the first step (DocumentParseAction) so that the document is just read once e.g.

final Validator validator = ObjectFactory.createValidator(scenarioType.getSchema());
validator.setErrorHandler(errorHandler);
validator.validate(new DOMSource(NodeOverNodeInfo.wrap(parseDocument.getUnderlyingNode())));

With this, I miss the line and column information. When using the byte-array from Input I get them. I'd think they are useful for debugging, so I wouldn't change this until this problem is solved. Any ideas?

@rkottmann what do you think?

from validator.

Since Saxon HE has no support for schema, things are more complicated. To run the schema validation, I need to re-read the source. Nevertheless, this reduces overall footprint and is implemented in current master branch.

When using an InputStream, the re-read requires a serialization prior validation. So using a stream is somewhat suboptimal:

• The stream is read when necessary
• must be serialized prior validation

Using an URI based Input is better, since the re-read is based on a new stream. Nevertheless, unless you are using the ByteArrayInput by using the corresponding read in InputFactory the target is not read into memory. It's directly parsed to XdmNode.

I also implemented support for using java.xml.transform.Source as an Input. The actual implemenation regarding streams is in SourceInput. In future version we may extend for other Source-Implementations as well.

@JannickWeisshaupt Feel free to test this implementation. Any help on spotting errors with this is welcome.

from validator.

Fixed with Release 1.3.0

from validator.