Motivation:
If the last item analyzed in a previous received HttpChunk/HttpContent was a part of an attribute's name, the read index was not set to the new right place and therefore raizing an exception in some case (since the "new" name analyzed is empty, which is not allowed so the exception).
What appears there is that the read index should be reset to the last valid position encountered whatever the case. Currently it was set when only when there is an attribute not already finished (name is ok, but content is possibly not).
Therefore the issue is that elements could be rescanned multiple times (including completed elements) and moreover some bad decoding can occur such as when in a middle of an attribute's name.
Modifications:
To fix this issue, since "firstpos" contains the last "valid" read index of the decoding (when finding a '&', '=', 'CR/LF'), we should add the setting of the read index for the following cases:
'lastchunk' encountered, therefore finishing the current buffer
any other cases than current attribute is not finished (name not found yet in particular)
So adding for this 2 cases:
undecodedChunk.readerIndex(firstpos);
Result:
Now the decoding is done once, content is added from chunk/content to chunk/content, name is decoded correctly even if in the middle of 2 chunks/contents.
A Junit test code was added: testChunkCorrect that should not raized any exception.
Motivation:
JdkZlibEncoder does not generate a GZIP header when a user does not write anything at all, ending up with a 10-byte footer only.
Modification:
Modify finishEncode() method to write the GZIP header if it was not written before.
Result:
Encoding 0-byte content result in 20-byte GZIP stream now.
Motivation:
Use simple retry mechanism when try to connect to peer so a slow-startup will not produce an error
Modification:
Add new Servlet params which allow to configure retry count and wait time. Default behaviour is the same as before
Result:
It is now possible to configure retry when the peer is not up yet and the Servlet is init.
Motivation:
CONTRIBUTING.md is useful only because it lets Github show a user the
link to it so the user can check what information we need before
submitting a bug report. However, Github does not do the same for a
pull request submission form, and thus there's no reason to keep the
information about how to submit a good pull request in CONTRIBUTING.md.
Modification:
Replace the section about issuing a pull request with the link to the
official developer guide.
Result:
CONTRIBUTING.md is easier to maintain.
Motivation:
We often receive a bug report or a pull request which do not give us
enough information. If CONTRIBUTING.md exists in the repository, Github
will display some notice in the beginning of the issue submission form,
which might increase the overall quality of the bug reports and pull
requests.
Modification:
Write CONTRIBUTING.md
Result:
Potentially higher-quality bug reports and pull requests
- Related issues: #1937#1938 and #1946
- Add InterfaceHttpPostRequestDecoder and Make HttpPostRequestDecoder
implement it
- HttpPostRequestDecoder actually delegates itself to
HttpPostStandardRequestDecoder or HttpPostMultipartRequestDecoder
- Deprecate IncompatibleDataDecoderException because it's not thrown
anywhere now
- Fixes#1950
- Use writtenBytes instead of localWrittenBytes so that operationProgressed() is triggered even when write operation was split into multiple system calls.
- This commit allows a user to write its first web socket frame right after calling WebSocketServerHandshaker.handshake() rather than adding a listener to the future it returns.
- Should fix#1933
