Prior to this patch the %f formatting of some FP values could be off by up
to 1, eg '%.0f' % 123 would return "122" (unix x64). Depending on the FP
precision (single vs double) certain numbers would format correctly, but
others wolud not. This patch should fix all cases of rounding for %f.
Float parsing (both single and double precision) may have a relative error
of order the floating point precision, so adjust tests to take this into
account by not printing all of the digits of the answer.
Prior to this patch, a float literal that was close to subnormal would
have a loss of precision when parsed. The worst case was something like
float('10000000000000000000e-326') which returned 0.0.
This patch improves parsing of floating point numbers by converting all the
digits (integer and fractional) together into a number 1 or greater, and
then applying the correct power of 10 at the very end. In particular the
multiple "multiply by 0.1" operations to build a fraction are now combined
together and applied at the same time as the exponent, at the very end.
This helps to retain precision during parsing of floats, and also includes
a check that the number doesn't overflow during the parsing. One benefit
is that a float will have the same value no matter where the decimal point
is located, eg 1.23 == 123e-2.
This patch changes how most of the plain math functions are implemented:
there are now two generic math wrapper functions that take a pointer to a
math function (like sin, cos) and perform the necessary conversion to and
from MicroPython types. This helps to reduce code size. The generic
functions can also check for math domain errors in a generic way, by
testing if the result is NaN or infinity combined with finite inputs.
The result is that, with this patch, all math functions now have full
domain error checking (even gamma and lgamma) and code size has decreased
for most ports. Code size changes in bytes for those with the math module
are:
unix x64: -432
unix nanbox: -792
stm32: -88
esp8266: +12
Tests are also added to check domain errors are handled correctly.
This returns a complex number, following CPython behaviour. For ports that
don't have complex numbers enabled this will raise a ValueError which gives
a fail-safe for scripts that were written assuming complex numbers exist.
IEEE floating point is specified such that a comparison of NaN with itself
returns false, and Python respects these semantics. This patch makes uPy
also have these semantics. The fix has a minor impact on the speed of the
object-equality fast-path, but that seems to be unavoidable and it's much
more important to have correct behaviour (especially in this case where
the wrong answer for nan==nan is silently returned).
These are now returned as "operation not supported" instead of raising
TypeError. In particular, this fixes equality for float vs incompatible
types, which now properly results in False instead of exception. This
also paves the road to support reverse operation (e.g. __radd__) with
float objects.
This is achieved by introducing mp_obj_get_float_maybe(), similar to
existing mp_obj_get_int_maybe().
This patch fixes 2 things when printing a floating-point number that
requires rounding up of the mantissa:
- retain the correct precision; eg 0.99 becomes 1.0, not 1.00
- if the exponent goes from -1 to 0 then render it as +0, not -0
I.e. they don't run successfully with MICROPY_LONGINT_IMPL_NONE
and MICROPY_LONGINT_IMPL_LONGLONG (the problem is that they generate
different output than CPython, TODO to fix that).
The use of large literal numbers is a big no-no when it comes to writing
programs which work with different int representations. Also, some checks
are pretty adhoc (e.g using struct module to check for 64-bitness). This
change bases entire detection on sys.maxsize and integer operarions, and
thus more correct, even if longer.
Note that this change doesn't mean that any of these tests can pass with
anything but MPZ - even despite checking for various int representations,
the tests aren't written to be portable among them.
C's printf will pad nan/inf differently to CPython. Our implementation
originally conformed to C, now it conforms to CPython's way.
Tests for this are also added in this patch.
This allows using (almost) the same code for printing floats everywhere,
removes the dependency on sprintf and uses just snprintf and
applies an msvc-specific fix for snprintf in a single place so
nan/inf are now printed correctly.
Previous to this patch, a big-int, float or imag constant was interned
(made into a qstr) and then parsed at runtime to create an object each
time it was needed. This is wasteful in RAM and not efficient. Now,
these constants are parsed straight away in the parser and turned into
objects. This allows constants with large numbers of digits (so
addresses issue #1103) and takes us a step closer to #722.
This fixes conversion when float type has more mantissa bits than small int,
and float value has small exponent. This is for example the case of 32-bit
platform using doubles, and converting value of time.time(). Conversion of
floats with larg exponnet is still not handled correctly.
acoshf, asinhf, atanhf were added from musl. mathsincos.c was
split up into its original, separate files (from newlibe-nano-2).
tan was added.
All of the important missing float functions are now implemented,
and pyboard now passes tests/float/math_fun.py (finally!).
See http://bugs.python.org/issue19995 for rationale. As micropython currently
aims for Python 3.3 compatibility I have not changed behaviour, but this
change allows the test suite to be run against a newer Python without having
spurious failures.
Tests in basics (which should probably be renamed to core) should not
rely on float, or import any non-built-in files. This way these tests
can be run when those features are not available.
All test in basics now pass on the pyboard using stmhal port, except for
string-repr which has some issues with character hex printing.
