Override protocol handling for http URLs
Metacat currently uses a socket system for uploading data files because of an
inefficiency in how http urls are handled by the default protocol handler. Need
to investigate overrriding that protocol handler so that this problem is not
Updated by Matt Jones over 22 years ago
There are concomitant changes in Morpho to produce these types of file uploads.
Morpho now uses the HTTPClient library for http communication, which
allows streaming data between client and server (Sun's http protocol
handler buffers all data before sending, which limits data transfer size
to memory available). No changes in metacat were needed to accomodate this
switch of protocol handlers on the client side; all changes were due to the
need for an alternative data encoding (other than URL encoding) that
efficiently supports binary files).
Modified Metacat to support large data file uploads. This is accomplished
by supporting a new content type for data sent to metacat:
which allows multiple files to be sent in a standard MIME format. The
MetacatServlet.handleGetOrPost() method now checks the incoming content
type and sends multipart/form-data encoded submissions to
MetacatServlet.handleMultipartForm(), which in turn delegates to
MetacatServlet.handleUploadAction(). When a client wants to load a data
file to metacat, now morpho creates a "multipart/form-data" encoded stream that
contains an "action" parameter set to the value "upload", a "docid"
parameter set to the value of the accession number for the file, and
a "datafile" part that contains the actual file contents. Because of a
bug / shortcoming in the com.oreilly.servlet library that I am using to parse
this mime encoding, the "action" and "docid" paramters MUST come before
the "datafile" part or metacat won't find the paramters.
The provided docid is inserted into xml_documents (after checking that it
meets expectations [unique, rev 1, etc]). Then the data file
recieved from the "upload" action is stored on the filesystem
using the Accession number provided in the docid. See
DocumentImpl.registerDocument for details.
This functionality completely replaces the socket-based functionality found
in DataFileUploadInterface and DataFileServer. It would still be nice to
abstract away the file storage functions so that they can be either written
to the filesystem or other locations (like the db ina BLOB) when they are
recieved in this new encoding format.
The cos.jar file is now needed for parsing the new encoding.