For some time, we've assumed that the URL specification had a mistake in
that it didn't allow the "#" character to appear decoded in the
fragment. We've gotten away with it so far.
However, turns out that the CoreFoundation NSURL class doesn't like it.
So we have to be stricter.
[ChangeLog][Important Behavior Changes][QUrl and QUrlQuery] QUrl no
longer decodes %23 found in the fragment to "#" in the output of
toString(QUrl::FullyEncoded) or toEncoded()
Task-number: QTBUG-31945
Change-Id: If5e0fb37bae84710986c9ca89bd69ec98437cd63
Reviewed-by: David Faure (KDE) <faure@kde.org>
It's a good practice to always replace bad UTF-8 sequences with the
replacement character. It could be considered a security issue too.
Change-Id: I9e7d72e4c4102cdb8334449b5e7f882228a9048f
Reviewed-by: David Faure (KDE) <faure@kde.org>
Those sections contain more than one components of a URL, separated by
delimiters. For that reason, QUrl::FullyDecoded and QUrl::DecodedMode do
not make sense, since they would cause the returned value to be
ambiguous and/or fail to parse again.
In fact, there was a comment in the test saying "look how it becomes
ambiguous".
Those modes are already forbidden in the setters and getters of the full
URL (setUrl(), url(), toString() and toEncoded()).
[ChangeLog][Important Behavior Changes][QUrl and QUrlQuery] QUrl no
longer supports QUrl::FullyDecoded mode in authority() and userInfo(),
nor QUrl::DecodedMode in setAuthority() and setUserInfo().
Change-Id: I538f7981a9f5a09f07d3879d31ccf6f0c8bfd940
Reviewed-by: David Faure (KDE) <faure@kde.org>
The longer explanation can be found in the comment in qurl.cpp. The
short version is as follows:
Up to now, we considered that every character could be replaced with
its percent-encoding equivalent and vice-versa, so long as the parsing
of the URL did not change. For example, x:/path+path and
x:/path%2Bpath were the same. However, to do this and yet be compliant
with most URL uses in the real world, we had to add exceptions:
- "/" and "%2F" were not the same in the path, despite the delimiter
being behind (rationale was the complex definition of path)
- "+" and "%2B" were not the same in the query, so we ended up not
transforming any sub-delim in the query at all
Now, we change our understanding based on the following line from
RFC 3986 section 2.2:
URIs that differ in the replacement of a reserved character with
its corresponding percent-encoded octet are not equivalent.
From now on, QUrl will not replace any sub-delim or gen-delim
("reserved character"), except where such a character could not exist
in the first place. This simplifies the code and removes all
exceptions.
As a side-effect, this has also changed the behaviour of the "{" and
"}" characters, which we previously allowed to remain decoded.
[ChangeLog][Important Behavior Changes][QUrl and QUrlQuery] QUrl no
longer considers all delimiter characters equivalent to their
percent-encoded forms. Now, both classes always keep all delimiters
exactly as they were in the original URL text.
[ChangeLog][Important Behavior Changes][QUrl and QUrlQuery] QUrl no
longer decodes %7B and %7D to "{" and "}" in the output of toString()
Task-number: QTBUG-31660
Change-Id: Iba0b5b31b269635ac2d0adb2bb0dfb74c139e08c
Reviewed-by: David Faure (KDE) <faure@kde.org>
On some platforms (like BlackBerry) qDebug doesn't write to stderr,
so we directly write to stderr with fprintf.
Change-Id: Ib86211c98cf4da1fa2dbea4600a78e2013dc1a5a
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
We don't know what it might be used for. The RFC for URI says it's an
HEXDIG, and since we uppercase all other HEXDIGs already (in
percent-encodings...).
Change-Id: I56d0a81315576dd98eaa2657c0307d79332543a5
Reviewed-by: David Faure (KDE) <faure@kde.org>
So far, this function hasn't been used for input coming in from the
user, so it wasn't necessary. But we may want to do it, or we may
already be doing it accidentally somewhere that isn't triggering the
failed assertions during unit testing.
So let's be on the safe side and allow it. And test it too.
Change-Id: Ib63addd8da468ad6908278d07a4829f1bdc26a07
Reviewed-by: David Faure (KDE) <faure@kde.org>
We have no idea what it might contain, but test it anyway to make sure
it works. Turns out there were a few bugs the unit tests have now
caught.
Change-Id: I0a6c868365feec31c2360b3c341c8ca6944f4352
Reviewed-by: David Faure (KDE) <faure@kde.org>
Registered names and IP addresses can only contain unreserved
characters (letters, digits, dots, hyphens, underscores) and the
colon, which is a gen-delim. For registered names and IPv4 addresses,
we can simply use the default config -- if anything that remains
percent-encoded, it means it's not a valid hostname anyway.
For IPv6, we just need to decode the colon.
Change-Id: If8083d47f6e5375f760e7a6c59631c89e4da8378
Reviewed-by: David Faure (KDE) <faure@kde.org>
The rule for a new override is that it must still work if the old
implementation is called. The catch is that any class that derives from
QProcess and isn't recompiled will still have QIODevice::open in its
virtual table. That is equivalent to overriding open() and calling
QIODevice::open() (like the tests).
In Qt 5.0, QProcess::start() called QIODevice::open directly, not the
virtual open(), so there's no expectation that a user-overridden open()
be called. With that in mind, simply fix QProcess::start to not call the
virtual open at all.
Similarly with QLocalSocket, the calls to open were always non-virtual.
Task-number: QTBUG-32284
Change-Id: I88925f0ba08bc23c849658b54582744997e69a4c
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
This is a bit like QDir::cleanPath(), but for URL paths.
The code is shared with QDir::cleanPath(), by extracting the common parts
it into a helper, qt_normalizePathSegments().
Change-Id: I7133c5e4aa2bf17fba98af13eb5371afba64197a
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This allows to find the parent directory url using
url.adjusted(QUrl::RemoveFilename).
Change-Id: I1ca433ac67e4f93080de54a9b7ab2e538509ed04
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
qt_ACE_do(".co.uk") was returning an empty string because of the
leading dot. Allow leading dots from topLevelDomain, but not from
other calls.
Change-Id: I757d9960708e205d30554cd2bbcf618c8624792b
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Calling waitForNotified on an uninitialized notifier will print a
warning and return false. The autotest has been adjusted.
Change-Id: I85e18d6d0a8a5462e1a5d451613add941d89b5fb
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
Do not lower case file names to generate hash keys since
QString::toLower() converts some characters with context
which the Windows file system will not.
Task-number: QTBUG-31341
Change-Id: I285bfedef3c1ca9d59083229e61974dd378c72ae
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@digia.com>
Instead of calling different wait functions on different platforms,
we use QTRY_COMPARE to check the process state.
Change-Id: I6489cabce9e63f9c8b1036f3cccbf35b52df72e7
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
Reviewed-by: Andreas Holzammer <andreas.holzammer@kdab.com>
This reverts commit e3fa266623b08e837cb4ccc7fe59da243d03dd27
That commit applied a change at the wrong place in the code.
Change-Id: I21e3045a3af14ad2f90c5fe338815c35a2d27ae6
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: David Faure (KDE) <faure@kde.org>
Leading and double dots are bad, but trailing dots are fine. The ASCII
part of a hostname is supposed to be LDH (letters, digits, hyphen) only,
but we accept '_' (underscore) as an exception too.
Change-Id: I79957ddec4da78a0e2357fe50c8687db03e1c99e
Reviewed-by: David Faure (KDE) <faure@kde.org>
qt_ACE_do(".co.uk") was returning an empty string because of the
leading dot. This has always caused issues in KDE code too, where ACE
normalization needs the dot removed, and re-added afterwards.
Change-Id: Id9fcea0333cf55c14d755a86d4bf33a50f194429
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@digia.com>
During a call to QDir::mkpath(), the same path could be created
by another process, in which case the OS mkdir will fail with EEXIST.
But the docs for mkpath() state that it's not an error if it
already exists, whereas for mkdir() it is an error. So
QFileSystemEngine::createDirectory should accept the EEXIST error
silently if it occurs while creating the sequence of parent directories
and the final leaf directory, but should fail if EEXIST happens when
it was called from QDir::mkdir(), which is when the createParents
parameter is false. We assume the operating system mkdir() and
CreateDirectory() are atomic, so there should be no race condition
in QDir::mkdir(). It's not necessary for mkpath() to call stat()
at each level, only to check whether an existing entry is a directory
or a file. Also added to the autotest to verify that if the
path is an existing file, creating a dir with the same name will
fail in either mkdir or mkpath.
Task-number: QTBUG-30046
Change-Id: I926352f10654fdf3b322c8685bb85ad8b8844874
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
qt_nameprep is tested by tst_qurlinternal. We just need to be sure that
QUrl handles them correctly.
Change-Id: Ic563004870d2cf2fa7a31ce49fff7280d5ffb5f3
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Edge case: a > that should have been >=. Without it, we never ran the
rest of the IDN nameprepping.
Change-Id: I2276d660de3a70d0c561bb18816820d9a0f47e77
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
RFC 3454 says about prohibited characters (section 2, "Preparation
Overview"):
3) Prohibit -- Check for any characters that are not allowed in the
output. If any are found, return an error. This is described in
section 5.
In other words, we mustn't simply strip the output of prohibited
characters. We must generate an error if they are present. We do that by
clearing the data.
We already had tests for prohibited output, but they were
indistinguishable from being stripped. So instead add some extra
characters so that we can tell whether the label was cleared.
Change-Id: I2d95217c27be5e2d54deed0036cb009e3b7f4886
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Because tests are usually run as root on some setups, it does not
make sense to test for the right permissions of a readonly file.
Change-Id: I484f88722d3a9ce7123edc0fb57acae528fa194e
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
In Qt 5, I managed to break the guarantee that a deserialised local
datetime is the same time of day (potentially different UTC time),
regardless of which timezone it was serialised in. This happened after
I fixed QTBUG-4057 with If650e7960dca7b6ab44b8233410a6369c41df73a,
which serialised datetimes as UTC.
This patch reverts QDateTime serialisation to pre-Qt 5 behaviour to
restore the guarantee and consequently re-opens QTBUG-4057.
Change-Id: Iea877f7ed886f530b928067789b53534e89fe8cb
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
If a process dies before all output is read into the internal buffer
of QProcess, we might lose data. Therefore we must drain the output
pipes like we already do in the synchronous wait functions.
Task-number: QTBUG-30843
Change-Id: I8bbc5265275c9ebd33218ba600267ae87d93ed61
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
Windows Embedded Compact does not have drive letters like desktop
Windows => do not try to build drive letter related test code for WEC7.
Change-Id: I2c3659220a001510c0555e2dd773b4dd68e9c2cc
Reviewed-by: Andreas Holzammer <andreas.holzammer@kdab.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
We don't need to keep an internal QBuffer position, we can just use the
one from QIODevice::pos(). It will keep track of goings ahead and
backwards for us, plus it will make the default bytesAvailable() work
out-of-the-box too.
This error was reported on IRC.
Change-Id: I8559e8ee56edaa01ca8732c1f1012082ebe3a3f2
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
Reviewed-by: Robin Burchell <robin+qt@viroteck.net>
The only way to make this possible is to disable the
atomic-rename-from-temp-file behavior. This is not done by default,
but only if the application allows this to happen.
https://bugs.kde.org/show_bug.cgi?id=312415
Change-Id: I71ce54ae1f7f50ab5e8379f04c0ede74ebe3136d
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
When QTemporaryFile attempts to create a new file in
createFileFromTemplate, it fails if the filename exists and is a
directory. Windows returns error code 5 (ERROR_ACCESS_DENIED) in this
case - rather than ERROR_FILE_EXISTS - which is not handled.
This patch handles ERROR_ACCESS_DENIED in addition to the already
handled ERROR_FILE_EXISTS, meaning that QTemporaryFile will continue
to look for unique names when a directory with the same name exists.
Task-number: QTBUG-30058
Change-Id: I42339887d7f5483e3dc6a03a9da15111c350da8f
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
Reviewed-by: David Faure <david.faure@kdab.com>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
OpenProcess + WaitForSingleObject is supposed to fail after the process exits,
but this seems to take some time until Windows notices.
Change-Id: I942a9b4a458c23fc4ac33b28386e28821128e991
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
Remove all trailing whitespace from the following list of files:
*.cpp *.h *.conf *.qdoc *.pro *.pri *.mm *.rc *.pl *.qps *.xpm *.txt *README
excluding 3rdparty, test-data and auto generated code.
Note A): the only non 3rdparty c++-files that still
have trailing whitespace after this change are:
* src/corelib/codecs/cp949codetbl_p.h
* src/corelib/codecs/qjpunicode.cpp
* src/corelib/codecs/qbig5codec.cpp
* src/corelib/xml/qxmlstream_p.h
* src/tools/qdoc/qmlparser/qqmljsgrammar.cpp
* src/tools/uic/ui4.cpp
* tests/auto/other/qtokenautomaton/tokenizers/*
* tests/benchmarks/corelib/tools/qstring/data.cpp
* util/lexgen/tokenizer.cpp
Note B): in about 30 files some overlapping 'leading tab' and
'TAB character in non-leading whitespace' issues have been fixed
to make the sanity bot happy. Plus some general ws-fixes here
and there as asked for during review.
Change-Id: Ia713113c34d82442d6ce4d93d8b1cf545075d11d
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
This broke the build on Android where fileno returns a short.
Change-Id: Ic8d32380078faeedcd22e785a912fede28251156
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
Add setProgram() and setArguments() methods to the QProcess api.
Add a convenient start(QIODevice::OpenMode) method.
Move the implementation of QProcess::start() to QProcess::open()
unifying the QProcess api with other QIODevice subclasses.
Change-Id: Id1af57da05f750fe8d526d391589c05ee8037bca
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Locking between processes, implemented with open(O_EXCL) on Unix
and CreateFile(CREATE_NEW) on Windows.
Supports detecting stale lock files and deleting them.
Advisory locking is used to prevent deletion of files that are still in use.
Change-Id: Id00ee2a4e77a29483d869037c7047c59cb909339
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Thiago Macieira
Bernd Weimer
David Faure
Frederik Gladhorn
David Faure
Sergio Ahumada
Joerg Bornemann
Friedemann Kleint
Shawn Rutledge
Rafael Roquetto
Mitch Curtis
Janne Anttila
Debao Zhang
Axel Waggershauser
Paul Olav Tvete
Corentin Jabot