In QCOMPARE, handle NaNs and infinities the way tests want them
handled, rather than by strict IEEE rules. In particular, if a test
expects NaN, this lets it treat that just like any other expected
value, despite NaN != NaN as float16 values. Likewise, format
infinities and NaNs specially in toString() so that they're reported
consistently.
Enable the qfloat16 tests that depend on this QCOMPARE() behavior.
Refise the testlib selftest's float test to test qfloat16 the same way
it tests float and double (and format the test the same way).
This is a follow-up to 37f617c405.
Change-Id: I433256a09b1657e6725d68d07c5f80d805bf586a
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The QTest::qCompare() implementations were almost duplicates; pull the
common code out into a templated version. Tweaked the
QTest::toString() specialization for float and double (a macro) and
fixed a bous modifier in double's format.
The doubleComparisons and floatComparisons tests in the tst_float.cpp
selftest shared a large block of tests in common, aside from the
difference of type. Break this out into a templated static function
to save duplication.
This prepares the way for using the same templated code for qfloat16.
Change-Id: I2823fd006910c5ff88335d625d1fa05cb7753513
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Tidied up the existing float tests in the process.
(In particular, s/SUCCESS/PASS/ since that matches real test output.)
These verify that QCOMPARE() handles floats and doubles as intended.
Extended the existing qFuzzyCompare tests to probe the boundaries of
the ranges of values of both types, in the process.
Revised the toString<double> that qCompare() uses to give enough
precision to actually show some of the differences being tested there
(12 digits, to match what qFuzzyCompare tests, so as to show different
values rather than, e.g. 1e12 for both expected and actual) and to
give consistent results for infinities and NaN (MinGW had eccentric
versions for these, leading to different output from tests, which thus
failed); did the latter also for toString<float> and fixed stray zeros
in MinGW's exponents (which made a kludge in tst_selftest.cpp
redundant, so I removed that, too).
That's further complicated handling of floating-point types, so let's
just keep an eye on how expensive that's getting by adding a benchmark
test for QTest::toString(). Unfortunately, default settings only get
runs that take modest numbers of milliseconds (some as low as 40)
while increasing this with -minumumvalue 100 or more gets the process
killed - and I'm unable to find out who's doing the killing (it's not
QProcess::kill, ::kill or the QtTest WatchDog, as far as I can tell).
So results are rather noisy; the integral tests exhibit speed-ups by
factors up to 5, and slow-downs by factors up to 100, between runs
with and without this change, which does not affec the integral tests.
The relatively modest slow-downs and speed-ups in the floating point
tests thus seem likely to be happenstance rather than signal.
Change-Id: I4a6bbbab6a43bf14a4089e96238a7c8da2c3127e
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
The Test Anything Protocol (TAP), was originally Perl's simple text-based
interface between testing modules and test harnesses, but has since been
adopted by a large number of producers and consumers in many different
languages, which allows colorizing and summarizing test results.
The format is very simple:
TAP version 13
ok 1 - test description
not ok 2 - test description
---
message: 'Failure message'
severity: fail
expected: 123
actual: 456
...
ok 3 - test description # SKIP
1..3
The specification [1] is very brief, so the implementation has been
based on how typical consumers behave, especially when it comes to
the undefined diagnostics block.
[1] http://testanything.org/tap-version-13-specification.html
Change-Id: I616e802ea380165c678510e940ddc6607d39c92d
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
