| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
| |
.. and breaking out of a loop where we should.
Reviewed-by: thiago
|
|
|
|
|
|
|
|
|
|
|
| |
If the file/data ends in the END CERTIFICATE line without a newline,
the certificate is still valid. If it's followed by anything other
than a newline, then it's no longer valid.
Also add another test for the BEGIN CERTIFICATE ending without
newline, to ensure we don't crash there either.
Reviewed-By: Peter Hartmann
|
|
|
|
| |
Reviewed-By: Peter Hartmann
|
|
|
|
| |
Reviewed-by: joao
|
|
|
|
|
|
|
|
| |
We had an hack in the code that chunked writes. Try to avoid
this since the hack is probably only needed for older windows versions.
Task-number: QTBUG-7344
Reviewed-by: Peter Hartmann
|
|
|
|
|
|
|
|
| |
It is now considered as finished immediatly after creation to avoid
an event loop spin.
Had to change the auto tests a bit to account for the new behaviour.
Reviewed-by: Peter Hartmann
|
|
|
|
|
|
|
| |
Thank you Matthew Cattell for the fix!
Task-number: QTBUG-6466
Reviewed-by: joao
|
|
|
|
| |
Reviewed-by: Markus Goetz
|
|
|
|
|
|
|
| |
We're using the ones from our parent class instead of own instance.
Reviewed-by: Thiago
Reviewed-by: joao
|
|
|
|
|
|
|
|
|
|
|
| |
The parsing would ignore the fact that "private" ended at the comma
and would instead try to get the continuation as the value.
We have to check if equal < comma and it's not -1. Using unsigned
comparisons does both things at once for us.
Task-number: QTBUG-7060
Reviewed-by: Peter Hartmann
|
|
|
|
| |
Reviewed-by: TrustMe
|
|
|
|
|
|
|
| |
Prefer a QHttpNetworkConnectionChannel which has a connected socket
to a channel which doesn't have a connected socket.
Reviewed-by: Peter Hartmann
|
|
|
|
| |
Reviewed-by: Peter Hartmann
|
|
|
|
|
|
|
|
|
|
| |
QHttpNetworkReply has now a throttled mode. If this is activated,
it will only read HTTP body when its internal buffer is empty.
This means that QNetworkReply.setReadBufferSize() can finally
be used for bandwidth limiting.
Thanks to David Faure for helping out.
Reviewed-By: Peter Hartmann
|
|
|
|
|
|
| |
Avoid calling socket->bytesAvailable().
Reviewed-by: Thiago
|
|
|
|
|
|
|
| |
This should get rid of QAbstractSocket warnings we sometimes emit.
Reviewed-by: Thiago
Reviewed-by: Peter Hartmann
|
|
|
|
| |
Reviewed-by: Paul
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
on the code path from QNetworkAccessManager::get() to
QTcpSocket::write() there were two calls involved that were invoked
via a QueuedConnection, which would in some occasions delay writing the
data to the socket. This patch makes sure the data is written to the
socket immediately, without returning to the event loop (only the HTTP
backend was changed for that event, the other backends were not
changed)
Reviewed-by: Thiago Macieira
Reviewed-by: Markus Goetz
|
|
|
|
|
|
|
|
| |
I moved the code above the other tests because it's more complex. But,
of course, that's wrong because it returns without setting the
necessary headers.
Reviewed-by: Peter Hartmann
|
|
|
|
|
|
| |
The max-age handling is already done in the cache-saving code.
Reviewed-by: Peter Hartmann
|
|
|
|
|
|
|
| |
PreferCache should not use the network even if must-revalidate is
set. But if it is not fresh enough, then it should use the network.
Reviewed-by: Peter Hartmann
|
|
|
|
| |
Reviewed-by: Peter Hartmann
|
|
|
|
| |
Reviewed-by: Trust Me
|
|
|
|
|
|
|
| |
There was a buffer that is always empty since it was only used for
parsing the HTTP proxy protocol, not the actual socket data.
Reviewed-by: Peter Hartmann
|
| |
|
|
|
|
|
|
| |
Let's see if this happens.
Reviewed-by: Thiago
|
|
|
|
|
|
|
|
| |
Instead of sequentially looking up and potentially taking
a long time, we now do work in parallel.
Reviewed-by: ogoffart
Reviewed-by: lars
|
|
|
|
| |
Fixes: QTBUG-6539
|
|
|
|
|
|
|
| |
This should finally enable us to use DirectConnection in
QNetworkReplyHandler of QtWebKit.
Reviewed-by: thiago
|
|
|
|
|
|
|
|
| |
If we have no string parsing in QDateTime, then we use sscanf, so we
must include <stdio.h>
Task-number: QTBUG-6668
Reviewed-by: Markus Goetz
|
|
|
|
|
|
|
|
|
|
|
|
| |
Actually, the RFC 2109 says cookies should not be stored if the path
attribute is not a prefix of the request URI the cookie comes from.
However, all browsers allow it anyway; with the demo browser e.g.
logging in to wordpress.com was not possible.
We still do path checking when sending cookies, i.e. in
QNetworkCookieJar::cookiesForUrl().
Reviewed-by: Markus Goetz
Task-number: QTBUG-5815
|
|
|
|
|
|
|
|
| |
Seems like WinHttpGetIEProxyConfigForCurrentUser can return more
than one proxy even if this is not documented.
Task-number: QTBUG-5981
Reviewed-by: Peter Hartmann
|
|
|
|
| |
Reviewed-by: Peter Hartmann
|
|
|
|
| |
Reviewed-by: TrustMe
|
|\ |
|
| |
| |
| |
| | |
Over src/ tools/ examples/ and demos/
|
|/
|
|
|
|
|
|
|
| |
The select() system call was used in the wrong way.
We need to select for exceptions too.
Task-number: QTBUG-5799
Reviewed-by: Aleksandar Sasha Babic <aleksandar.babic@nokia.com>
Reviewed-by: Peter Hartmann
|
|
|
|
|
|
| |
Don't use QFile's Qt3support member, use fileName() instead.
Reviewed-by: Trust me
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The QFileNetworkReply is a wrapper around QFile that
has therefore similar performance. This avoids
the usage of the unperformant QNetworkAccessFileBackend.
The benchmark qfile_vs_qnetworkaccessmanager shows
that the QFileNetworkReply's performance is better
than 0.9x of QFile compared to QNetworkAccessFileBackend
which had about 0.5x of QFile.
Reviewed-by: Peter Hartmann
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The QDateTimeParser forces at least two QString allocations
to convert the Month string into a integer. This makes parsing
the date string the most expesnsive operation from within
replyHeaderChanged.
Use sscanf to parse the RFC1123 header to get a significant
speedup. Use a switch case statement to convert the month
name to a integer.
Reviewed-By: Markus Goetz
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
toLower will force a memory allocation, and memory allocations
are certainly adding a delay. Use qstricmp were this is easily
possible. This change makes headerValues et all drop out of
the profile.
QChar::toLower will now be called more often, specially for
loops that do linear searching. In the profile with real web
content this is faster than using QByteArry::toLower in the
inner loop and the forced memory allocations.
Reviewed-By: Markus Goetz
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Using qstricmp is faster on the desktop than the alloc of
QByteArray::toLower and the QByteArray::operator==.
QChar::toLower will now be called for the input key from
within the loop. In the measurements done this is still
a performance improvement over the allocation done by
QByteArray::toLower. For more header values this might
not be the case anymore.
RESULT : tst_Loading::byteArrayTestLower():
0.00076 msec per iteration (total: 25, iterations: 32768)
RESULT : tst_Loading::byteArrayTestCompare():
0.00028 msec per iteration (total: 37, iterations: 131072)
Reviewed-By: Markus Goetz
|
| |
| |
| |
| |
| |
| |
| | |
HTTP header names are case insensitive, but stick to what all browsers
etc. use to work around buggy servers.
Reviewed-by: TrustMe
|
| |\ |
|
| | |
| | |
| | |
| | |
| | |
| | | |
Inspired by merge request 2012 by Joel Nordell
Reviewed-by: Peter Hartmann
|
| | |
| | |
| | |
| | | |
Reviewed-by: Peter Hartmann
|
| |/
|/|
| |
| |
| |
| |
| |
| | |
HTTP header names are case insensitive, but stick to what all browsers
etc. use to work around buggy servers.
Task-number: QTBUG-5618
Reviewed-by: Peter Hartmann
|
|\ \ |
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
instead than when parsing the cookie header.
This corrects the bug QT-2379, happening in the following sequence:
parseCookie -> setCookieUrl -> toRawForm -> parseCookie
where a default domain would now also have a dot prefix, and shouldn't.
QT-2379
Reviewed-by: Peter Hartmann
|