This updates `dtoa.c` to avoid using the Bigint free-list in --disable-gil builds and
to pre-computes the needed powers of 5 during interpreter initialization.
* gh-111962: Make dtoa thread-safe in `--disable-gil` builds.
This avoids using the Bigint free-list in `--disable-gil` builds
and pre-computes the needed powers of 5 during interpreter initialization.
* Fix size of cached powers of 5 array.
We need the powers of 5 up to 5**512 because we only jump straight to
underflow when the exponent is less than -512 (or larger than 308).
* Rename Py_NOGIL to Py_GIL_DISABLED
* Changes from review
* Fix assertion placement
This PR removes `_Py_dg_stdnan` and `_Py_dg_infinity` in favour of
using the standard `NAN` and `INFINITY` macros provided by C99.
This change has the side-effect of fixing a bug on MIPS where the
hard-coded value used by `_Py_dg_stdnan` gave a signalling NaN
rather than a quiet NaN.
---------
Co-authored-by: Mark Dickinson <dickinsm@gmail.com>
Remove the HAVE_PY_SET_53BIT_PRECISION macro (moved to the internal
C API).
* Move HAVE_PY_SET_53BIT_PRECISION macro to pycore_pymath.h.
* Replace PY_NO_SHORT_FLOAT_REPR macro with _PY_SHORT_FLOAT_REPR
macro which is always defined. gcc -Wundef emits a warning when
using _PY_SHORT_FLOAT_REPR but the macro is not defined, if
pycore_pymath.h include was forgotten.
* Fix failure of _Py_dg_dtoa to remove trailing zeros
* Add regression test and news entry
* Add explanation about why it's safe to strip trailing zeros
* Make code safer, clean up comments, add change note at top of file
* Nitpick: avoid implicit int-to-float conversion in tests
Move the dtoa.h header file to the internal C API as pycore_dtoa.h:
it only contains private functions (prefixed by "_Py").
The math and cmath modules must now be compiled with the
Py_BUILD_CORE macro defined.
Fix an unlikely memory leak on conversion from string to float in the
function _Py_dg_strtod() used by float(str), complex(str),
pickle.load(), marshal.load(), etc.
Fix an unlikely memory leak in _Py_dg_strtod() on "undfl:" label:
rewrite memory management in this function to always release all
memory before exiting the function. Initialize variables to NULL, and
set them to NULL after calling Bfree() at the "cont:" label.
Note: Bfree(NULL) is well defined: it does nothing.
svn+ssh://pythondev@svn.python.org/python/trunk
........
r77691 | mark.dickinson | 2010-01-22 16:18:09 +0000 (Fri, 22 Jan 2010) | 1 line
Correct typo in comment.
........
r77698 | mark.dickinson | 2010-01-22 17:04:07 +0000 (Fri, 22 Jan 2010) | 3 lines
Issue #7743: Fix a potential incorrect rounding bug in dtoa.c (2nd bug
in issue 7743).
........
r77713 | mark.dickinson | 2010-01-23 20:48:56 +0000 (Sat, 23 Jan 2010) | 3 lines
Issue #7743: Add checks for zero inputs to the lshift and mult functions;
this fixes the first bug described in issue #7743.
........
r77714 | mark.dickinson | 2010-01-23 21:25:53 +0000 (Sat, 23 Jan 2010) | 1 line
dtoa.c fix from upstream that fixes incorrectly rounded results for certain subnormals that are also halfway cases.
........
svn+ssh://pythondev@svn.python.org/python/trunk
........
r77589 | mark.dickinson | 2010-01-17 20:57:56 +0000 (Sun, 17 Jan 2010) | 7 lines
Issue #7632: When Py_USING_MEMORY_DEBUGGER is defined, disable the
private memory allocation scheme in dtoa.c, along with a piece of code
that caches powers of 5 for future use. This makes it easier to
detect dtoa.c memory leaks with Valgrind or similar tools.
Patch by Stefan Krah.
........
svn+ssh://pythondev@svn.python.org/python/trunk
........
r77519 | mark.dickinson | 2010-01-16 10:44:00 +0000 (Sat, 16 Jan 2010) | 5 lines
Issue #7632: Fix a serious wrong output bug for string -> float conversion.
Also remove some now unused variables, and add comments clarifying the
possible outputs of the parsing section of _Py_dg_strtod. Thanks
Eric Smith for reviewing.
........
r77530 | mark.dickinson | 2010-01-16 17:57:49 +0000 (Sat, 16 Jan 2010) | 3 lines
Issue #7632: Fix one more case of incorrect rounding for str -> float
conversion (see bug 5 in the issue tracker).
........
r77533 | mark.dickinson | 2010-01-16 18:06:17 +0000 (Sat, 16 Jan 2010) | 1 line
Fix multiple uses of variable 'L' in _Py_dg_strtod, where one use requires an unsigned long and the other a signed long. See also r77421.
........
svn+ssh://pythondev@svn.python.org/python/trunk
........
r77477 | mark.dickinson | 2010-01-13 18:21:53 +0000 (Wed, 13 Jan 2010) | 1 line
Add comments explaining the role of the bigcomp function in dtoa.c.
........
r77478 | mark.dickinson | 2010-01-13 19:02:37 +0000 (Wed, 13 Jan 2010) | 1 line
Clarify that sulp expects a nonnegative input, but that +0.0 is fine.
........
r77481 | mark.dickinson | 2010-01-13 20:55:03 +0000 (Wed, 13 Jan 2010) | 1 line
Simplify and annotate the bigcomp function, removing unused special cases.
........
r77482 | mark.dickinson | 2010-01-13 22:15:53 +0000 (Wed, 13 Jan 2010) | 1 line
Fix buggy comparison: LHS of comparison was being treated as unsigned.
........
r77483 | mark.dickinson | 2010-01-13 22:20:10 +0000 (Wed, 13 Jan 2010) | 1 line
More dtoa.c cleanup; remove the need for bc.dplen, bc.dp0 and bc.dp1.
........
r77490 | mark.dickinson | 2010-01-14 13:02:36 +0000 (Thu, 14 Jan 2010) | 1 line
Fix off-by-one error introduced in r77483. I have a test for this, but it currently fails due to a different dtoa.c bug; I'll add the test once that bug is fixed.
........
r77491 | mark.dickinson | 2010-01-14 13:14:49 +0000 (Thu, 14 Jan 2010) | 1 line
Issue 7632: fix a dtoa.c bug (bug 6) causing incorrect rounding. Tests to follow.
........
r77492 | mark.dickinson | 2010-01-14 14:40:20 +0000 (Thu, 14 Jan 2010) | 1 line
Issue 7632: fix incorrect rounding for long input strings with values very close to a power of 2. (See Bug 4 in the tracker discussion.)
........
r77493 | mark.dickinson | 2010-01-14 15:22:33 +0000 (Thu, 14 Jan 2010) | 1 line
Issue #7632: add tests for bugs fixed so far.
........
svn+ssh://pythondev@svn.python.org/python/trunk
........
r77410 | mark.dickinson | 2010-01-10 13:06:31 +0000 (Sun, 10 Jan 2010) | 1 line
Remove unused BCinfo fields and an unused macro.
........
r77421 | mark.dickinson | 2010-01-11 17:15:13 +0000 (Mon, 11 Jan 2010) | 1 line
Change a variable type to avoid signed overflow; replace repeated '19999' constant by a define.
........
r77450 | mark.dickinson | 2010-01-12 22:23:56 +0000 (Tue, 12 Jan 2010) | 4 lines
Issue #7632: Fix a problem with _Py_dg_strtod that could lead to
crashes in debug builds, for certain long numeric strings
corresponding to subnormal values.
........
r77451 | mark.dickinson | 2010-01-12 22:55:51 +0000 (Tue, 12 Jan 2010) | 2 lines
Issue #7632: Fix a bug in dtoa.c that could lead to incorrectly-rounded results.
........
- incorporate and adapt David Gay's dtoa and strtod
into the Python core
- on platforms where we can use Gay's code (almost
all!), repr(float) is based on the shortest
sequence of decimal digits that rounds correctly.
- add sys.float_repr_style attribute to indicate
whether we're using Gay's code or not
- add autoconf magic to detect and enable SSE2
instructions on x86/gcc
- slight change to repr and str: repr switches
to exponential notation at 1e16 instead of
1e17, str switches at 1e11 instead of 1e12
Steve (Gadget) Barnes
Sam Gross
Victor Stinner
Sebastian Berg
Eric Snow
junyixie
Mark Dickinson
Victor Stinner
Stefan Krah
Benjamin Peterson
Serhiy Storchaka
Victor Stinner
Mark Dickinson
Brett Cannon