If _Thread_local is used on a block-scope declaration, it must be
combined with either static or extern to decide linkage.
Change-Id: I228b3520767197c6cdf5134ff5a666ab2aca33ea
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Commit 29bc68cf16 added support for
unsigned and commit 5ff7a3d96e later added
support for int. This commit adds support for qsizetype, which isn't int
on 64-bit platforms.
We do this by reorganizing the code and using the generic version of
__builtin_{add,sub,mul}_overflow from GCC 5 and Clang 3.8, which ICC 18
seems to support now too on Linux. That leaves older versions of GCC and
Clang, as well as MSVC, ICC on Windows, and the GHS compiler, to use the
generic implementations, as I've removed the assembly code those
versions of GCC and Clang on x86 are now uncommon.
Note: any older version of ICC probably breaks. We only support the
latest.
Change-Id: I9e2892cb6c374e93bcb7fffd14fc11bcd5f067a7
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
Also drops a few instances where the dependency was purely runtime,
especially for examples.
Change-Id: I2a0476f79928143596bdb3b8f01193af90574ae8
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
Unary ~ is not defined for enum classes, so we need a cast.
Change-Id: I79d495ebcc24ab960da8dae3be08eb307a9de448
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Both ARM and x86 can convert fp16 much faster in bulk than one at a
time. This also enables hardware accelerated conversion on x86, when
F16C isn't unconditionally available at compile time.
This code is implemented in C to ensure that there's no leakage of
inline symbols from the .obj file that was compiled by Visual Studio
with AVX support. Unfortunately, simd.prf uses $(CXX) instead of $(CC)
for all its sources, which means the file gets interpreted as C++ by
g++, clang++ and icpc. Those compilers at least don't leak any symbols.
Done-with: Thiago Macieira <thiago.macieira@intel.com>
Change-Id: I9d26d99e83392861fb09564e0e8e8d76cd8483b3
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Tested with Clang, GCC 4.5 & up, ICC 17 and MSVC 2017. No current
version of MSVC supports C11 and GCC implemented the features slightly
later in C than in C++.
Change-Id: I57a1bd6e0c194530b732fffd14f45c5074c9a052
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
This brings us to almost parity with the C++11 Random Engine API
requirements (see chapter 26.5.1.4 [rand.req.eng]). We don't implement
the templated Sseq requirements because it would require moving the
implementation details to the public API. And we don't implement the
<iostreams> code because we don't want to.
Change-Id: Icaa86fc7b54d4b368c0efffd14f05ff813ebd759
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Since we don't document how many bytes one needs (it's 2496), it's
difficult for the caller to provide just enough data in the seed
sequence. Moreover, since std::mt19937 doesn't make it easy to provide
the ideal size either, we can't actually write code that operates
optimally given a quint32 range either -- we only provide it via
std::seed_seq, which is inefficient.
However, we can do it internally by passing QRandomGenerator to the
std::mersenne_twister_engine constructor, as it's designed to work.
Change-Id: Icaa86fc7b54d4b368c0efffd14f0613c10998321
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Now only QRandomGenerator::system() will access the system-wide RNG,
which we document to be cryptographically-safe and possibly backed by a
true HWRNG. Everything else just wraps a Mersenne Twister.
Change-Id: I0a103569c81b4711a649fffd14ec8cd3469425df
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Since we're adding a deterministic generator that inherently does not
use syscalls, and people should really use that one by default, there is
no point in optimizing the secure generator wrt syscalls. Besides,
keeping the random data in memory for longer than needed is likely
inadviseable.
Change-Id: Ib17dde1a1dbb49a7bba8fffd14ed0871117fe930
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Right now,this does really nothing. This commit is just to allow us to
transition the other modules (besides qtbase) to use the syntax that
will become the API.
I've marked three places to use the system CSPRNG:
1) the QHash seed
2) QUuid
3) QAuthenticator
I didn't think the HTTP multipart boundary needed to be
cryptographically safe, so I changed that one to the global generator.
Change-Id: Ib17dde1a1dbb49a7bba8fffd14ecf1938bd8ff61
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
The previous version was good, just not optimal. Because the input was
an unsigned 64-bit number, compilers needed to generate extra code to
deal with HW instructions that only convert 64-bit signed input. And
that was useless because a double uniformly distributed from 0 to 1 can
only have 53 bits of randomness.
The previous implementation did exactly what the Microsoft libstdc++ and
libc++ implementations do. In my opinion, those implementations have an
imperfect distribution, which is corrected in this commit. In those, all
random input bigger than 0x20000000000000 has a different frequency
compared to input below that mark. For example, both 0x20000000000000
and 0x20000000000001 produce the same result (4.8828125e-4).
What's more, for the libc++ and MSVC implementations, input between
0xfffffffffffff001 and 0xffffffffffffffff results in 1.0 (probability 1
in 2⁵³), even though the Standard is very clear that the result should
be strictly less than 1. GCC 7's libstdc++ doesn't have this issue,
whereas the versions before would enter an infinite loop.
Change-Id: Ib17dde1a1dbb49a7bba8fffd14eced3c375dd2ec
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Error out if it's missing or broken (Mersenne Twister not present).
This ensures that we never have a low-quality random generator in Qt.
Change-Id: I0a103569c81b4711a649fffd14ec80649df7087e
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
"generate" is better than "get", and we already have "generate(it, it)"
which uses std::generate(). This changes:
- get32() → generate()
- get64() → generate64() and QRandomGenerator64::generate()
- getReal() → generateDouble()
Change-Id: I6e1fe42ae4b742a7b811fffd14e5d7bd69abcdb3
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
[ChangeLog][QtCore] Added qEnvironmentVariable, which returns the value
of an environment variable in a QString, while qgetenv continues to be
used to return it in a QByteArray. For Unix, since most environment
variables seem to contain path names, qEnvironmentVariable will do the
same as QFile::decodeName, which means NFC/NFD conversion on Apple OSes.
I opted not to #include <qfile.h> from qglobal.cpp to implement that
QFile::decodeName functionality, so qglobal.cpp doesn't depend on
corelib/io and to avoid possible recursions.
Task-number: QTBUG-41006
Change-Id: I14839ba5678944c2864bffff141794b8aaa7aa28
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Found while working on suppressing the warning about the return value
(which is either 0 or -1) was being ignored.
Task-number: QTBUG-61968
Change-Id: I02d22222fff64d4dbda4fffd14d148b1724547ca
Reviewed-by: Florian Bruhin <qt-project.org@the-compiler.org>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The getentropy function, first found in OpenBSD, is present in glibc
since version 2.25 and Bionic since Android 6.0 and NDK r11. It uses the
Linux 3.17 getrandom system call. Unlike glibc's getrandom() wrapper,
the glibc implementation of getentropy() function is not a POSIX thread
cancellation point, so we prefer to use that even though we have to
break the reading into 256-byte blocks.
The big advantage is that these functions work even in the absence of a
/dev/urandom device node, in addition to a few cycles shaved off by not
having to open a file descriptor and close it at exit. What's more, the
glibc implementation blocks until entropy is available on early boot, so
we don't have to worry about a failure mode. The Bionic implementation
will fall back by itself to /dev/urandom and, failing that, gathering
entropy from elsewhere in the system in a way it cannot fail either.
uClibc has a wrapper to getrandom(2) but no getentropy(3). MUSL has
neither.
Change-Id: Ia53158e207a94bf49489fffd14c8cee1b968a619
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Makes the qle_bitfield template more generic and moves it to qendian_p.h
It is also hardened to be more reliable.
Change-Id: I53214ec99cceee4f5e8934ae688c99e555a5fb42
Reviewed-by: Ville Voutilainen <ville.voutilainen@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
GCC didn't support it until version 5 or 6, so add configure tests for
both <random> and <sys/auxv.h>. Normally I'd say "upgrade", but this is
too low-level and important a feature.
There's a good chance that all our supported compilers have <random>
anyway. As for <sys/auxv.h>, it's present on Glibc, Bionic and MUSL, but
I don't see it in uClibc (AT_RANDOM is a Linux-specific feature).
Change-Id: Ia3e896da908f42939148fffd14c5b2af491f7a77
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
This class provides a reasonably-secure random number generator that
does not need seeding. That is quite unlike qrand(), which requires a
seed and is low-quality (definitely not secure).
This class is also like std::random_device, but better. It provides an
operator() like std::random_device, but unlike that, it also provides a
way to fill a buffer with random data, not just one 32-bit quantity.
It's also stateless.
Finally, it also implements std::seed_seq-like generate(). It obeys the
standard requirement of the range (32-bit) but not that of the algorithm
(if you wanted that, you'd use std::seed_seq itself). Instead,
generate() fills with pure random data.
Change-Id: Icd0e0d4b27cb4e5eb892fffd14b4e3ba9ea04da8
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
The documentation says that it's equivalent to
qgetenv(varName).toInt()
But the implementation wasn't. QByteArray::toInt() verifies that the
entire string was consumed, so QByteArray("1a").toInt() == 0, but
qstrtoll alone doesn't. That is, qstrtoll("1a", ...) == 1.
The implementation also detected the base, a behavior I kept. Instead, I
updated the documentation.
Change-Id: I0031aa609e714ae983c3fffd14676ea6061a9268
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Backtrace logging tests were not passing for arm when -O2 option was used.
Set "-fno-inline" on for the app whose backtrace is to be inspected.
Task-number: QTBUG-59966
Change-Id: Id1bbf78c31dc524357a30c7d39c239689621b155
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
The operator double() and operator long double() members of qfloat16
are causing cast ambiguities. This removes them, leaving only
operator float() which seems to be adequate.
Also, additional arithmetic operator tests were added which without
this removal fail to compile.
Change-Id: Id52a101b318fd754969b3de13c1e528d0aac2387
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
I don't know why.
tst_qflags.cpp(114): error: no instance of function template "verifyConstExpr" matches the argument list argument types are: (Qt::MouseButton)
tst_qflags.cpp(91): note: this candidate was rejected because there is a type mismatch after argument substitution
Change-Id: I84e363d735b443cb9beefffd14b9581d77933cb8
Reviewed-by: Tony Sarajärvi <tony.sarajarvi@qt.io>
Backtrace logging tests were not passing for arm. Added compile option
-funwind-tables to support backtrace on arm.
Task-number: QTBUG-59966
Change-Id: I5e2443b1e3a644a239dab68db990e75ae8fade24
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Replace all QT_NO_PROCESS with QT_CONFIG(process), define it in
qconfig-bootstrapped.h, add QT_REQUIRE_CONFIG(process) to the qprocess
headers, exclude the sources from compilation when switched off, guard
header inclusions in places where compilation without QProcess seems
supported, drop some unused includes, and fix some tests that were
apparently designed to work with QT_NO_PROCESS but failed to.
Change-Id: Ieceea2504dea6fdf43b81c7c6b65c547b01b9714
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
This constitutes a fairly complete submission of an entirely new
floating point type which conforms to IEEE 754 as a 16-bit storage
class. Conversion between qfloat16 and float is currently performed
through a sequence of lookup tables. Global-level functions
qRound(), qRound64(), qFuzzyCompare(), qFuzzyIsNull(), and
qIsNull() each with a qfloat16 parameter have been included
for completeness.
[ChangeLog][QtCore] Added new qfloat16 class.
Change-Id: Ia52eb27846965c14f8140c00faf5ba33c9443976
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
So that QFlags can use an (un)signed int matching the
underlying type as identified by the compiler and not by us.
Requires fixing a few warnings about sign conversion due to
QFlags misusages in qtbase that were either plain wrong, or
were relying on the enum being backed by an (un)signed int
when it wasn't.
Keep qtypetraits.h in the source tree in order to prevent
source breaks if some downstream #includes it (note however
that it did not contain any public API).
Change-Id: Ib3a92b98db7031e793a088fb2a3b306eff4d7a3c
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This new function does the same as newRow(), except that it has a less confusing
name (in line with _add_Column()), and accepts printf-style arguments to avoid
the need to newRow(qPrintable(QString::asprintf())), a common pattern in client
code. It uses qvsnprintf() under the hoods, avoiding the need for the QString
const char* round-trip.
Port all in-tree users of newRow(qPrintable(QString::asnprintf())) to the new
function.
Change-Id: Icd5de9b7ea4f6759d98080ec30f5aecadb8bec39
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Namely: decltype(). Ideally we'd want C++17's template constructor
argument deduction, but instead use the C++11 solution: a factory
function. This enables using things such as lambdas in the container
argument.
Change-Id: Idba64d8069d15bbafe54cfdebe24b1fba1eb8d0a
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Chaining hooks together was mentioned by Ossi in the comments of
d953d9a4. This patch justs add a test that verifies that it works, and
also serves as an informal example for developers looking how to do it.
Change-Id: I53a014d5663c289ea0559e0926ed301f4e5110e6
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
Reviewed-by: hjk <hjk@qt.io>
Any other use than for enums should use std::is_[un]signed. Make this
explicit by renaming the type traits.
Change-Id: I494158563c95c710e710d0d337f4e547006df171
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Since the macro is now just a wrapper for std::is_enum,
its use is also deprecated.
[ChangeLog][QtCore][Global] Q_IS_ENUM is deprecated.
Use std::is_enum<>::value instead.
Change-Id: I09b9f4559c02c81f338cace927873318f2acafde
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Remove most type traits from qtypetraits.h, but keep the custom
implementation of is_signed/is_unsigned. This gets rid of
BSD-3 licensed code from Google in a public header (hugh!).
The custom implementations for is_signed/is_unsigned are kept
because the implementations in gcc's standard headers do not
work as we expect for enums - both is_signed and is_unsigned
always returns false there - see also
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59027
[ChangeLog][QtCore][General] Qt now relies on type traits from
the C++ standard library.
Change-Id: I3f2188b46949f04ca4482a6ac9afd3482103f0e1
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Consistent with other Unix platforms, and internally consistent between tests,
as a lot of tests were already applying CONFIG -= app_bundle manually.
Change-Id: Icd2b7e1c08015b26137af60ff82fddbc753f0ff4
Reviewed-by: Jake Petroules <jake.petroules@qt.io>
