Uwe Ryssel created CXF-7684:
-------------------------------
Summary: Base64 encoding in AttachmentSerializer does not create
correct output for large attachments
Key: CXF-7684
URL: https://issues.apache.org/jira/browse/CXF-7684
Project: CXF
Issue Type: Bug
Components: Core
Affects Versions: 3.0.16
Reporter: Uwe Ryssel
As client, I want to send a SOAP message with base64-encoded attachments. I
have activated base64 encoding for attachments by adding an out interceptor,
which sets the message parameter
org.apache.cxf.message.Message.CONTENT_TRANSFER_ENCODING to "base64", since it
is supported by the CXF-core's AttachmentSerializer.
For small attachments, the encoding works. But when I have a large attachment,
the base64-encoded data is, I think, not valid:
As implemented in AttachmentSerializer's method encodeBase64(), the encoding of
the input data is done in blocks of maximum 262144 bytes. Since 262144 divided
by 3 has remainder 1, each of the blocks containing the full 262144 bytes will
generate two '=' padding characters at the end of the block output.
So for large attachments with > 262144 bytes, the created base64 data stream
contains padding characters within the stream, and not only at the end. I did
not find any hint, that this is allowed.
Especially the server implementation, which I use, interprets the padding
characters also as end of stream, so that attachments > 262144 do not work in
that case.
An fix for this issue would be to use a buffer size, which is divisible by
three without remainder.
I still use CXF 3.0.16, since the client have to work with Java 1.6. But 3.1.x
and 3.2.x use the same method, so the same issue should apply there too.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
