Decimal.canonical() returns a modified argument (when it says it doesn't)

[tanant]

Bug report

Bug description:

According to the docs for 3.11.11 (and up)

Return the canonical encoding of the argument. Currently, the encoding of a Decimal instance is always canonical, so this operation returns its argument unchanged.

However this doesn't seem to be the case:

for example in ["10", 
                "10.2", 
                "-3.2e2",     # missing the + 
                "3.2e1",      # possibly redundant e-notation
                "3.2e2",      #
                "-9.2e-1",    # -ve exponent
                "-9.2e-6",    # -ve exponent
                "-9.23e-6",    # -ve exponent
                ]:
    print("{:>10} : {:>10}".format(example, decimal.Decimal(example).canonical()))

#         10 :         10
#       10.2 :       10.2
#     -3.2e2 :    -3.2E+2  # understandable token change to E+
#      3.2e1 :         32  # unexpected
#      3.2e2 :     3.2E+2  # as per argument, but what's special about e+2 and up?
#    -9.2e-1 :      -0.92  # really unexpected
#    -9.2e-6 : -0.0000092  # really unexpected
#    -9.2e-7 :    -9.2E-7  # unsure why -7 and greater flips behaviour

The token change sort of makes sense but the other cases seems somewhat at odds with the docstring

Context

I'm doing some work in a 3rd party JSON dump where most of the numbers are in scientific notation. I need to read in the file, modify some non-numerical strings, and then dump everything back out. I know the values are numerically equivalent but downstream processes are doing checks to ensure that the only changes in the JSON are literally the expected changes, which are text-string only modifications.

I was writing up checks and was wondering why my re-encoder was throwing a fit, and saw this in my test suite.

CPython versions tested on:

3.11

Operating systems tested on:

Windows

[JelleZijlstra]

You're printing strings in your first column, not Decimal instances. What's modifying these objects is not the .canonical() call, but the Decimal constructor. .canonical really does return its argument unchanged:

cpython/Modules/_decimal/_decimal.c

Line 4475 in ba2d2fd

dec_canonical(PyObject *self, PyObject *Py_UNUSED(dummy))

[tanant]

Ahhh, so the constructor is changing the value, not .canonical() within Decimal itself.. riiight, that makes sense now you point it out. That's definitely a nuance I didn't think of, thanks!

[JelleZijlstra]

The constructor isn't exactly changing the value either, it parses a string into a Decimal object, and multiple strings may parse to the same Decimal object.

[tanant]

Yep, I've sort of rejigged my head into considering the problem I'm solving slightly differently as something more like a limited text parser, feels like a more appropriate model here.