v0.2.2

This release contains #121, making it so the poisson_binom gufunc kernels in stats.h initialize outputs to contain all zeros, rather than making it a requirement for the caller to pass in zeroed out arrays for the outputs.

v0.2.1

This release contains #117, adding additional gufunc kernels for use in scipy.stats.poisson_binom to improve the efficiency of pmf and cdf calculation.

v0.2.0

This release adds the following new functions:

• #97 adding the asymptotic (infinite sample size) null distribution CDF of the single-sample Cramér-von Mises statistic
• #97 adding the null distribution CDF of the single-sample Cramér-von Mises statistic.
• #113 adding the pmf of the Binomial Poisson distribution.

the following improvements and bug fixes for existing functions:

• #85 fixing a special case of hyp2f1.
• #86 adding missing pointer dereference in an asymptotic expansion used in cyl_bessel_i.
• #95 adding missing NaN checks in hyp2f1.
• #102 replacing atan(y/x) with atan2(y, x) whenever possible.

and the following general maintenance work

• #100 cleaning up incorrect names for constants defined in cephes/const.h and adjusting some of the values to be correctly rounded.
• #103 adding operators around a boolean expression in fp_error_metrics.h to avoid a compiler warning.
• #106 adding an equivalent of np.linspace for use in test code.
• #114 removing a variable that became unused after an earlier PR in this release.

Contributors to this release by PR count were
@fbourgey 6
@steppi 3
@maffoo 1
@stratakis 1

Thanks everyone!

v0.1.5

This is a bug fix release. It contains the following fixes:

1. #80 by @fbourgey , fixing the n = 0 case of lqn by using the correct limiting formula.
2. #81 by @WarrenWeckesser, updating code translated from AMOS to consistently use the std:: qualifier.
3. #82 by @fbourgey , fixing the m = 0, x = 1, v non-integer case of lpmv by using a correct limiting formula.
4. #83 by @fbourgey, fixing the large t case of itj0y0 by correcting a mistranslation of the original Fortran.

Thanks @fbourgey and @WarrenWeckesser!

v0.1.4

This is a bug fix release. It contains the following fixes:

• MAINT: fix maybe-uninitialized compiler warning in amos.h #75 by @rgommers, which gets rid of a maybe-unitialized compiler warning that appears with GCC 13.3.0 in SciPy's CI by initializer a variable when it is defined that previously was only being initialized in some conditional code paths.
• Bugfix/Workaround sign error Asso. Legendre Function at z=-1 #76 by @mberz and @steppi fixes assoc_legendre_p in the case when z = -1, m=0, and n is odd.
• BUG: Fix mistranslated expression in amos.h #69 by @WarrenWeckesser fixes some edge cases in cyl_bessel_k
• fix: prevent array out-of-bounds write in segv function.Add boundary … #74 by @Inuyasha-fan adds a bounds check to prevent a segfault in segv used in spheroidal wave functions from specfun.
• MAINT: change [project] to [workspace] in pixi.toml. #73 by @WarrenWeckesser removes a deprecated field in pixi.toml.
• MAINT: In numpy.h, define two more function types: llld_d and qqqd_d #77 by @WarrenWeckesser adds new function types to numpy.h which are needed in SciPy.

Thanks @Inuyasha-fan, @mberz, @rgommers, @WarrenWeckesser!

v0.1.3

This is a bug fix release. It contains the following fixes:

#51 from @SabrinaJewson replaces uses of sf_error, which is defined in SciPy's codebase but not xsf's, with the correct set_error, helping ensure xsf compiles when used outside of SciPy.
#58 from @Emiliano-DM fixes amos.h to use the C++ header <complex> instead of the C header <complex.h>, fixing compilation errors that arise due to the macro #define complex _Complex defined in <complex.h>.
In #62, @jorenham, fixed incorrect bracket placement in lqmn which caused derivatives to not be computed for some values of z. This incorrect bracket placement was likely introduced at some point in the translation of this code into C++.
#63 from jorenham updates mdspan to the latest stable version, fixing some build failures that were appearing on Windows.
In #66, @ChrisAB fixed expi to handle the special case x = inf correctly.

In addition, #54 from @SabrinaJewson fixes a typo in the README, #64 from @jorenham cleans up a unused test file which was left in inadvertently, and #42 from @lucascolley adds a CI job to check formatting with clang-format.

Thanks @ChrisAB, @Emiliano-DM, @jorenham, @lucascolley, for your contributions, and thanks @WarrenWeckesser for helping to review #58.

v0.1.2

This is a bug fix release. It adds some missing execution space annotations (with the XSF_HOST_DEVICE macro) for functions which are not currently used in CuPy, but which are needed when compiling functions which are used in CuPy when using CUDA toolkit 11.

v0.1.1

This release makes minor changes needed to make xsf work as a submodule of CuPy. There are no changes to the (currently undocumented) public API. It includes the pull request, #39. GPU tests have also been added through CuPy.

v0.1.0

This is an initial release mostly for internal use in SciPy. The code is not yet documented or organized, and the structure reflects the historical development of the scalar kernel codebase of SciPy special. Most of this code is fairly well tested, though there are known defects that still need to be addressed. Feel free to try it out, but expect breaking changes.