Issue #983 | Add Round function and implement the conversion to C, Fortran, and Python

[nandiniraja348]

I have added the Round Python function.

Added tests.

[EmilyBourne]

Hi, how are you doing? Are you still working on this PR? Do you need a hand? I just wanted to remind you to keep your branch up to date with the master branch. There have been a few aesthetic changes recently (mainly PR #1012) which should be merged sooner rather than later to avoid accumulating conflicts

[EmilyBourne]

round in python has a very strange way of handling values that are equally close to 2 integers (see here):

For the built-in types supporting round(), values are rounded to the closest multiple of 10 to the power minus ndigits; if two multiples are equally close, rounding is done toward the even choice (so, for example, both round(0.5) and round(-0.5) are 0, and round(1.5) is 2). Any integer value is valid for ndigits (positive, zero, or negative). The return value is an integer if ndigits is omitted or None. Otherwise, the return value has the same type as number.

This needs to be handled, maybe we need to implement our own round function in pyc_math for c and fortran

[aihya]

An error is raised when I compile the following code for Fortran:

if __name__ == "__main__":
    print(round(42, ndigits=5))

The error:

ERROR at shared library generation stage
Traceback (most recent call last):
  File "/home/glitch/Documents/nandiniraja348/nandiniraja348_env/bin/pyccel", line 33, in <module>
    sys.exit(load_entry_point('pyccel', 'console_scripts', 'pyccel')())
  File "/home/glitch/Documents/nandiniraja348/pyccel/commands/console.py", line 241, in pyccel
    execute_pyccel(filename,
  File "/home/glitch/Documents/nandiniraja348/pyccel/codegen/pipeline.py", line 382, in execute_pyccel
    generated_program_filepath = src_compiler.compile_program(compile_obj=prog_obj,
  File "/home/glitch/Documents/nandiniraja348/pyccel/codegen/compiling/compilers.py", line 345, in compile_program
    self.run_command(cmd, verbose)
  File "/home/glitch/Documents/nandiniraja348/pyccel/codegen/compiling/compilers.py", line 429, in run_command
    raise RuntimeError(err_msg)
RuntimeError: Failed to build module
/home/glitch/Documents/nandiniraja348/__pyccel__/prog_test_round.f90:9:36:

    9 |   write(*, '(F0.12)', advance="yes") pyc_bankers_round(42_i64, 5_i64)
      |                                    1
Error: There is no specific function for the generic ‘pyc_bankers_round’ at (1)

[yguclu]

@aihya Good catch! This is because the two functions implemented in Fortran require the first argument to be a float (single or double precision), while here you are passing an integer.

@nandiniraja348 Unit tests should be added to cover the case of integer input.

cc: @EmilyBourne

[aihya]

Following code is compiled for C

if __name__ == "__main__":
    print(round(42, ndigits=4.2))

Output:

$> pyccel test_round.py --language c
/home/glitch/Documents/nandiniraja348/pyccel/codegen/compiling/compilers.py:431: UserWarning: /home/glitch/Documents/nandiniraja348/__pyccel__/prog_test_round.c: In function ‘main’:
/home/glitch/Documents/nandiniraja348/__pyccel__/prog_test_round.c:7:15: warning: format ‘%ld’ expects argument of type ‘long int’, but argument 2 has type ‘double’ [-Wformat=]
    7 |     printf("%ld\n", lrint(42 * pow(10.0, 4.2)) / pow(10.0, 4.2));
      |             ~~^     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
      |               |                                |
      |               long int                         double
      |             %f

  warnings.warn(UserWarning(err))

$> ./test_round
140733447113224

Incorrect rounded value when passing non-integer to ndigits:

if __name__ == "__main__":
    print(round(4.2, ndigits=4.2))

Output: 4.200030659267

[EmilyBourne]

@aihya
Pyccel doesn't translate code which doesn't run in python. Your example:

if __name__ == "__main__":
    print(round(4.2, ndigits=4.2))

gives:

TypeError                                 Traceback (most recent call last)
<ipython-input-1-dd8329dcb3e7> in <module>
----> 1 print(round(4.2, ndigits=4.2))

TypeError: 'float' object cannot be interpreted as an integer

[EmilyBourne]

The last failing test is due to an integer not being stored accurately in a floating point. I'm not sure how to fix this without using python's integer round method (operating on strings). Does anyone have any ideas?