Changes implemented for asynchronous loading in the browser with Saxon-JS 2.0:
-
ixsl:schedule-action/@document can now also be used to asynchronously load other documents (e.g. text, and JSON), not just XML documents
- internally we now first store the HTTP response in a
responsePool, and then process the content only at the point that a resource retrieval function is actually called (e.g. fn:doc, fn:document-available, fn:json-doc, fn:unparsed-text, etc.); at which point the processed result is stored in one of two pools (documentPool for XML documents, or textResourcePool for other resources)
Some further internal implementation details:
When an asynchronous resource load is called (using ixsl:schedule-action/@document) internally, in the callback for the async "GET" XMLHttpRequest, rather than attempting to process the HTTP response directly (to put the result in the documentPool, as in Saxon-JS 1), we now store the response in the responsePool. The key for an entry is the absolute URI of the resource, and the value will either be an XError (if the fetch failed), or a new JSON object, with properties responseText (whose value is the responseText string from the HTTP response) and contentType (from the "Content-Type" header of the HTTP response).
Subsequently, when any of the resource retrieval functions is called (fn:doc, fn:document-available, fn:json-doc, fn:unparsed-text, etc.), we obtain the relevant response from the response pool, and now process the content as required. For instance, checking for invalid XML characters for the document functions and unparsed-text, but not for json-doc; and parsing as XML for the document functions. The result for the document functions is stored in the documentPool, and the result for the unparsed-text functions (including json-doc) is stored in the textResourcePool. (The intention is that the contentType response information may be used for handling encoding, e.g. when supplied in the 2nd arg of unparsed-text, but this is not yet fully working.)
These pools, and the response processing methods, are of course also used for synchronous loads (i.e. those from the internal platform.readFile method). Note that synchronous loads are discouraged in the browser. Also note that we currently only support synchronous loads in Saxon-JS 2 on the Node.js platform, for loading resources from file system, using the fs module.
As well as tidying up the three main fetch methods for XML and JSON documents, and unparsed text (now called fetchDoc, fetchJsonDoc, and fetchUnparsedText for clarity, and found in the new Fetch.js module); the methods for loading of other resources has also been tidied up (to try to eliminate unnecessary duplication, improve clarity, etc.); e.g. for fetching the source XML and SEF before initiating a transform.
Changes committed on the trunk 2.0 development branch: main changes are in BrowserPlatform.js, NodeJSPlatform.js, CoreFn.js, transform.js, and new module Fetch.js (which gathers relevant methods together, some previously in CoreFn.js e.g. fetchDoc, which didn't really properly belong there).
.