diff options
author | Edward Welbourne <edward.welbourne@qt.io> | 2020-04-15 13:10:31 +0200 |
---|---|---|
committer | Edward Welbourne <edward.welbourne@qt.io> | 2020-04-30 13:44:08 +0200 |
commit | c78a960198d59fb9a9ddd83ad7098b1db396edfd (patch) | |
tree | 3e8eec0b57cd67796743ebb56fa883ca24ee2f69 /src/testlib | |
parent | 91d18c646443c812addf6d1cdada99d3c2d3a3ec (diff) |
QCOMPARE: treat values as equal if qFuzzyIsNull(each)
We hope this shall avoid some flaky failures noticed in quick tests,
e.g. tst_QQuickMenu::Material::subMenuPosition(cascading,flip) was
recently seen failing with 3.88e-11 != 0. This required some revision
to test data in the testlib selftest for floats; the resulting
expected output differs in details but not in which tests pass or
fail. QEMU, naturally, made life difficult, requiring special-case
code in the test-driver.
[ChangeLog][QtTestLib][QCOMPARE] QCOMPARE() now treats its values as
equal when qFuzzyIsNull() is true for both of them.
Change-Id: Icc6ad5164b609937eddbe39cc69120f0abf0f3b4
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Mitch Curtis <mitch.curtis@qt.io>
Diffstat (limited to 'src/testlib')
-rw-r--r-- | src/testlib/qtestcase.cpp | 17 | ||||
-rw-r--r-- | src/testlib/qtestcase.qdoc | 21 |
2 files changed, 21 insertions, 17 deletions
diff --git a/src/testlib/qtestcase.cpp b/src/testlib/qtestcase.cpp index f45999c7fa..74507c11e1 100644 --- a/src/testlib/qtestcase.cpp +++ b/src/testlib/qtestcase.cpp @@ -1,6 +1,6 @@ /**************************************************************************** ** -** Copyright (C) 2019 The Qt Company Ltd. +** Copyright (C) 2020 The Qt Company Ltd. ** Copyright (C) 2016 Intel Corporation. ** Contact: https://www.qt.io/licensing/ ** @@ -2525,16 +2525,21 @@ bool QTest::compare_helper(bool success, const char *failureMsg, } template <typename T> -static bool floatingCompare(const T &t1, const T &t2) +static bool floatingCompare(const T &actual, const T &expected) { - switch (qFpClassify(t1)) + switch (qFpClassify(expected)) { case FP_INFINITE: - return (t1 < 0) == (t2 < 0) && qFpClassify(t2) == FP_INFINITE; + return (expected < 0) == (actual < 0) && qFpClassify(actual) == FP_INFINITE; case FP_NAN: - return qFpClassify(t2) == FP_NAN; + return qFpClassify(actual) == FP_NAN; default: - return qFuzzyCompare(t1, t2); + if (!qFuzzyIsNull(expected)) + return qFuzzyCompare(actual, expected); + Q_FALLTHROUGH(); + case FP_SUBNORMAL: // subnormal is always fuzzily null + case FP_ZERO: + return qFuzzyIsNull(actual); } } diff --git a/src/testlib/qtestcase.qdoc b/src/testlib/qtestcase.qdoc index e4e1825bb5..72f8cdaf8c 100644 --- a/src/testlib/qtestcase.qdoc +++ b/src/testlib/qtestcase.qdoc @@ -1,6 +1,6 @@ /**************************************************************************** ** -** Copyright (C) 2019 The Qt Company Ltd. +** Copyright (C) 2020 The Qt Company Ltd. ** Contact: https://www.qt.io/licensing/ ** ** This file is part of the documentation of the Qt Toolkit. @@ -109,19 +109,18 @@ continues. If not, a failure is recorded in the test log and the test function returns without attempting any later checks. - Always respect QCOMPARE() parameter semantics. The first parameter passed to it - should always be the actual value produced by the code-under-test, while the - second parameter should always be the expected value. When the values don't - match, QCOMPARE() prints them with the labels \e Actual and \e Expected. - If the parameter order is swapped, debugging a failing test can be confusing. + Always respect QCOMPARE() parameter semantics. The first parameter passed to + it should always be the actual value produced by the code-under-test, while + the second parameter should always be the expected value. When the values + don't match, QCOMPARE() prints them with the labels \e Actual and \e + Expected. If the parameter order is swapped, debugging a failing test can be + confusing and tests expecting zero may fail due to rounding errors. When comparing floating-point types (\c float, \c double, and \c qfloat16), - \l qFuzzyCompare() is used for finite values. Infinities match if they have + \l qFuzzyCompare() is used for finite values. If qFuzzyIsNull() is true for + both values, they are also considered equal. Infinities match if they have the same sign, and any NaN as actual value matches with any NaN as expected - value (even though NaN != NaN, even when they're identical). This means that - expecting 0 can fail when the actual value may be affected by rounding errors. - One solution to this is to offset both actual and expected values by adding - some suitable constant (such as 1). + value (even though NaN != NaN, even when they're identical). QCOMPARE() tries to output the contents of the values if the comparison fails, so it is visible from the test log why the comparison failed. |