After:

 CC="gcc -DPY_NO_SHORT_FLOAT_REPR" ./configure && make

test_format_testfile in test_float fails with:

test.support.TestFailed: Traceback (most recent call last):
  File "Lib/test/test_float.py", line 341, in test_format_testfile
    self.assertEqual(fmt % float(arg), rhs)
AssertionError: '100000000000.0' != '1e+11'

The problem is that for str(float), the new formatting code switches to
exponential notation at 1e11, not 1e12; the legacy code does not.

A similar issue exists for repr: repr(1e16) should now produce '1e+16'.

Here's a patch. Eric, do you have time to check this over for general
sanity?

On second thoughts, I think that repr shouldn't change: 'legacy'
repr format should mean that it *is* exactly the same as before,
including the switch at 1e17 instead of 1e16. Otherwise there will be 3
possibilities for repr output: 'short' repr, 'legacy' repr and pre-3.1
repr will all be different.

Here's a new patch, which leaves repr alone and fixes up the Misc/NEWS
entries accordingly.

I agree that there should be 2 behaviors, not 3.

This patch looks okay to me, and tests pass both with and without
defining PY_NO_SHORT_FLOAT_REPR.

One nit: You don't need to use Py_CHARMASK when using the ISDIGIT macro
locally defined in pystrtod.c. On the other hand, that macro (ISDIGIT)
should completely go away. I think it's there because this file is
trying to be locale-unaware, but isdigit from ctype.h is defined as
being locale-unaware (at least on all platforms I have access to and a
search on the net; my copy of C89 is at home and I can't triple-check
just yet).

Also, remove_trailing_zeros should go inside the #ifdef so it's not
included if Gay's code is being used.

I seem to recall that despite the C standards, there are platforms around
where isdigit and isxdigit are still locale aware. This bit me when I
implemented float.fromhex: see r65964.

And my man page for isdigit says:

COMPATIBILITY
The 4.4BSD extension of accepting arguments outside of the range of
the unsigned char type in locales with large character sets
is considered obsolete and may not be supported in future releases.
The iswdigit() function should be used instead.

So I'm tempted to go the other way and say that all uses of isdigit should
be replaced by ISDIGIT.

I'll fix the Py_CHARMASK and move remove_trailing_zeros inside the ifdef.

I definitely think that if you go with isdigit, you need the macro. I've
been bitten by this in 3.0.

My actual suggestion is to go with isdigit and the macro everywhere in
this file and all files.

But for just this checkin, ISDIGIT is probably best.

Committed, with the fixups Eric pointed out, in r71692.