ChangeLog

==================
e-antic Change Log
==================

.. current developments

v2.0.2
====================

**Fixed:**

* Fixed exception being thrown by PyPy due to https://github.com/wlav/cppyy/issues/209.



v2.0.1
====================

**Fixed:**

* Fixed segfault when using non-intrusive pointers to ``renf_class`` such as the ones returned by ``parent()`` from Python.
* Fixed compilation with GCC 13.2.1.



v2.0.0
====================

**Added:**

* Added tarball with the built documentation to our GitHub release pages with every future release of e-antic.
* Added support for FLINT 3.

**Removed:**

* Removed the static library from the conda package that we upload with every commit to the master branch to the flatsurf channel. To our knowledge nobody is using these builds. They can sometimes be useful for debugging because they are built with debug symbols (which is why they are very big.) Providing a static library that nobody should use anyway is eating up the limited storage on anaconda.org twice as fast so we disable this.

**Fixed:**

* Adapted CI setup on GitHub to changes in setup-miniconda.
* Fixed some typos in source code, documentation, and error messages.
* Improved `environment.yml` files for development to not pull in any incompatible dependencies from PyPI silently
* Fixed compiler warning about possibly uninitialized variable.



v1.3.0
====================

**Added:**

* Added support for FLINT 2.8 and 2.9
* Added short and unsigned short operators to C++ interface.



v1.2.3
====================

**Added:**

* Added automatic update on conda-forge with every release.

**Fixed:**

* Fixed soname which was not bumped in the 1.2.2 release.



v1.2.2
====================

**Fixed:**

* Fixed conda channels used for binder notebooks.
* Fixed compilation errors on recent compilers by upgrading catch2 dependency used for testing.
* Fixed deserialization of renf_class from old Python pickles. (#242 sometimes crashed when called from Python since cppyy had trouble with the exceptions being thrown.)
* Fixed "installation with conda" page in our documentation (#240)
* Added logo to documentation. (#239)
* Fixed serialization with cereal 1.3.1+.
* Fixed compiler warnings about missing prototypes for test functions.



v1.2.1
====================

**Fixed:**

* Fixed serialization of `renf_class` with cereal>=1.3.2 (#224)

* Fixed deserialization of objects serialized with e-antic prior to 1.0.0



v1.2.0
====================

**Performance:**

* Improved conversion from vectors of rational coefficients to `renf_elem_class` by using a faster code path in FLINT.

* Improved conversion from SageMath number fields to pyeantic `RealEmbeddedNumberField`. This **adds a dependency** of pyeantic on gmpxxyy.



v1.1.1
====================

**Deprecated:**

* Deprecated arithmetic between elements in different number fields in the C++ interface. Before, an operation such as `a + b` was possible even if `a` and `b` lived in different number fields if at least one of them was actually a rational number. This led to inconsistent parent fields of the resulting element. We still allow such operations if the parent of `a` or `b` is the rational field (`renf_class::make()`) otherwise a deprecation warning is printed. To fully opt in to the new behaviour set the environment variable `LIBEANTIC_STRICT_BINOP` to any value; this raises an exception instead of printing a warning.

**Fixed:**

* Fixed a compilation error on some xenial systems.
* Fixed import order in pyeantic's setup.py script.
* Fixed inclusion of .map files in distribution tarball even if building without version script support.

**Performance:**

* Improved performance of `RealEmbeddedNumberField` in Python interface by caching results. In particular the costly conversion from `renf_class`.



v1.1.0
====================

**Added:**

* automatic integration checks for [Normaliz](https://github.com/Normaliz/Normaliz)
* Added support for Python 3.10; we are now automatically testing pyeantic with
  Python 3.10.
* Added `renf_class` that uses a root with dynamic precision. (Makes it easier to specify a root when the roots of the minimal polynomial are very close to each other.)

**Changed:**

* We do not claim anymore that `_cmp` functions return -1, 0, 1. Instead we
  only make a statement about the sign of the integer returned. In practice we
  still return -1, 0, 1 for the time being.
* Header file `e-antic.h` is now called `local.h`. There is still a header file
  `e-antic.h` that includes all of e-antic's C interface, so in particular
  `local.h`.
* Header file `renfxx_fwd.hpp` is now called `forward.hpp`. The old header file
  `renfxx_fwd.hpp` can still be used.
* Header file `renfxx.h` is now called `e-antic.hpp`. The old header file
  `renfxx.h` can still be used.
* Dropped explicit support for Python 3.6 which has reached its end of life. It
  should still work but we are not testing it explicitly in our CI anymore.

**Fixed:**

* We do not assume anymore that `_cmp` functions return -1, 0, 1. FLINT, GMP,
  and Arb only guarantee that these functions return a negative, zero, or
  positive integer. In practice their implementations return -1, 0, 1, however
  in some obscure case on Fedora/i686, this was not the case.
* Removed unused includes that cannot be resolved on Windows.
* Fixed a compiler warning on Fedora.
* Improved the error message when arb headers cannot be found. Fedora installs the arb headers into /usr/include/arb instead of /usr/include.
* Added missing `LIBEANTIC_API` declarators so building with `-fvisibility=hidden` produces a usable libeanticxx.so.
* Updated installation instructions for installation from conda-forge. The
  e-antic package there has been renamed to libeantic. The package `e-antic`
  still exists but is now essentially an alias for `libeantic` and `pyeantic`.
* Worked around doctesting issues in SageMath 9.4.
* Fixed failing SageMath tests when pytest is installed.
* Fixed construction of `RealEmbeddedNumberField` in Python when roots of the minimal polynomial are very close to each other.



v1.0.3
====================

**Fixed:**

* Cleaned up the README and our Zenodo site.

* Fixed linking of C++ test programs.



v1.0.2
====================

**Fixed:**

* removed docbuild warnings for latest version of mkdocs.
* Fixed parsing of trivial number fields in pyeantic. #197
* Conversion from `NumberField` to `RealEmbeddedNumberField` when defining polynomial is not in `x`.



v1.0.1
====================

**Fixed:**

* `./configure` now checks for cppyythonizations which are required to run Python tests
* allow creation of `RealEmbeddedNumberField` from an intrusive pointer in pyeantic.
* Skip tests requiring realalg and sage when configured --without-realalg and --without-sage, respectively.
* Disable pyeantic testing --without-pytest.



v1.0.0
====================

**Added:**

* Added `renf_class::construction()` which produces the parameters that can be used to create a `renf_class` with `renf_class::make()`.
* Run CI on macOS & Linux
* Added `renf_class::get_pword` to extract the number field set with `renf_class::set_pword` to ease migration of legacy code.
* Added `.map` files so only API symbols get exported when running `./configure --with-version-script`.
* Added visibility attributes to header files; this will be necessary on Windows and allows compilation with `-fvisibility=hidden`
* Conda packages are now uploaded automatically to the flatsurf channel with every release.
* Integrated the Python wrapper pyeantic into the e-antic repository
* Added `renf_elem_class::floordiv()` to perform `(a/b).floor()` more efficiently
* Floor division of `renf_elem_t`
* Added the missing implementations of free `floor`, `ceil`, and `pow` functions to the C++ library.
* Added `renf_elem_swap(renf_elem_t, renf_elem_t)`
* There is now `e-antic/renfxx_fwd.hpp` if you only need forward declarations
  of `renf_class` and `renf_elem_class`.
* `renf_elem_class` can now be created from signed and unsigned long long.
* `renf_elem_class` can now be created from vectors of primitive integers, e.g.,
  ```
  renf_elem_class x(K, {1, 2, 3}); // = 3*x^2 + 2*x + 1
  ```
  where before the entries of the vector had to be `mpz_class`.
* Move semantics `&&` have been added to `renf_elem_class`.
* There is now support for serialization with cereal. See t-cereal.cpp for
  examples on how to use it.
* The release process has been automated with [rever](https://regro.github.io/rever-docs/)

**Changed:**

* Split `renfxx.h` into `renf_elem_class.hpp` and `renf_class.hpp` headers. The
  original header still exists and simply includes the two novel headers.
* Split the header `e-antic/poly_extra.h` into `e-antic/fmpz_poly_extra.h` and
  `e-antic/fmpq_poly_extra.h`. Typically, you do not want to include either of
  those as they are mostly meant to be used internally.
* C++ Comparison operators of `renf_elem_class` are now free so they participate in
  implicit casts in the same way like the boost generated free operators that
  are derived from them.
* The C++ interface now requires C++14.
* All C++ classes are now declared in the namespace `eantic`.
* The semantics of the C++ `operator =` have changed. In e-antic 0.1 the
  following would create the unit in the field K.
  ```
  renf_elem_class x(K);
  x = 1;
  ```
  Now the above statement makes x a rational number. More generally, an
  assignment resets the number field so that after `x = y` the condition
  `x.nf == renf_elem_class(y).nf` holds. To mimic the old behavior you need
  to rewrite the above as
  ```
  renf_elem_class x(K);
  x = renf_elem_class(K, 1);
  ```
* The C++ `renf_class` is now hidden behind a factory to get smart pointer
  semantics everywhere. Create a `renf_class` by calling `renf_class::make(…)`.
  This returns a smart pointer, so you might have to replace some `.` with
  `->`.
* The change of semantics in assignment also affects reading from streams (in
  order to create `renf_elem_class`). Before the following would parse an element
  into a number field:
  ```
  renf_elem_class x(K);
  in >> x;
  ```
  Now this only works if the stream contains a rational number. (Otherwise an
  exception is raised.) As `in >> x` also resets `x.nf`. The above code should
  be replaced with:
  ```
  renf_elem_class x;
  K.set_pword(in);
  in >> x;
  ```
* `string renf_class::gen_name` is now a method so it needs to be called.
* Many operations that threw an exception before when domains were mixed, now
  abort program execution (typically through a call to `assert()`.) You are not
  supposed to mix domains unless explicitly stated otherwise.
* `renf_class.operator==` now also checks that the generator name is the same.
  Similarly, `renf_class.operator=` now also resets the generator name.
* Changed renf_elem generator in C++ unit tests to always start with the zero
  element since it probably does not get generated by chance and can obviously
  cause a lot of trouble.

**Deprecated:**

* Some methods have been deprecated and might be removed in a future release,
  mostly to make the interface more consistent. The deprecation warnings give
  hints which methods to use instead.

**Removed:**

* `renf_elem_class(string&)` has been removed. If you want to parse a rational
  number, use `renf_elem_class(mpq_class(string))`. If you want to parse into a
  number field, use `renf_elem_class(renf_class&, string&)`.
* `renf_elem_class::operator=(string&)` has been removed. If you want to parse
  a rational into an element, use `x = mpq_class(string)`. If you want to parse
  into a number field, use `x = renf_elem_class(x.parent(), string)`.
* `renf_elem_class(vector<...>)` have been removed as it would have thrown an
  exception always anyway.
* `renf_elem_class::operator=(vector<...>)` have been removed due to the change
  of semantics of `=`. If `x` is not a rational you get the same behaviour as
  before with `x = renf_elem_class(x.parent(), {1, 2, 3})`.
* `renf_class::xalloc()` has been removed and replaced by an implementation
  detail.

**Fixed:**

* Compatible with FLINT 2.7
* Removed dependency of `renfxx_cereal.h` on Arb and FLINT, i.e., one does not need to `-larb -lflint` anymore when using cereal.
* Set pins in Conda packages as they are in https://github.com/conda-forge/conda-forge-pinning-feedstock
* `E_ANTIC_RELEASE` now reports the correct version number, i.e., 10000.
* e-antic had claimed to be thread safe via an open MP pragma (in the number
  field refinement). In some cases, there was a problem with thread-safety. We
  now require users to explicitly mark multithreaded sections by forbidding
  mutations to a renf, see `renf_set_immutable`. As a result, there are some
  operations that cannot be done anymore in a multi-threaded environment but
  they now fail properly (instead of leading to random crashes.)

**Performance:**

* Speed up arithmetic between `renf_elem_class` in C++ interface.
* Speed up creation of trivial number fields by a factor of seven.
* Speed up `renf_class::operator==`



v1.0.0-rc.16
====================

**Fixed:**

* Fixed arithmetic with rationals in different number fields.



v1.0.0-rc.15
====================

**Fixed:**

* Made lexical_cast include work with more versions of boost.



v1.0.0-rc.14
====================

**Changed:**

* Split `renfxx.h` into `renf_elem_class.hpp` and `renf_class.hpp` headers. The
  original header still exists and simply includes the two novel headers.

**Performance:**

* Speed up arithmetic between `renf_elem_class` in C++ interface.
* Speed up creation of trivial number fields by a factor of seven.



v1.0.0-rc.13
====================

**Added:**

* Added `renf_class::construction()` which produces the parameters that can be used to create a `renf_class` with `renf_class::make()`.

**Fixed:**

* compatible with FLINT 2.7
* Removed dependency of `renfxx_cereal.h` on Arb and FLINT, i.e., one does not need to `-larb -lflint` anymore when using cereal.



v1.0.0-rc.12
====================

**Added:**

* run CI on macOS & Linux
* added `renf_class::get_pword` to extract the number field set with `renf_class::set_pword` to ease migration of legacy code.
* added `.map` files so only API symbols get exported when running `./configure --with-version-script`.
+ added visibility attributes to header files; this will be necessary on Windows and allows compilation with `-fvisibility=hidden`

**Fixed:**

* set pins in Conda packages as they are in https://github.com/conda-forge/conda-forge-pinning-feedstock
* adapt to changes in cppyy 1.9.0
* `E_ANTIC_RELEASE` now reports the correct version number, i.e., 10000.



v1.0.0-rc.11
====================

**Fixed:**

* fixed pyeantic version bump in rever script



v1.0.0-rc.10
====================

**Added:**

* Conda packages are now uploaded automatically to the flatsurf channel with every release.
* integrated pyeantic into the e-antic repository



v1.0.0-rc.9
====================

**Changed:**

* Split the header `e-antic/poly_extra.h` into `e-antic/fmpz_poly_extra.h` and
  `e-antic/fmpq_poly_extra.h`. Typically, you do not want to include either of
  those as they are mostly meant to be used internally.

**Fixed:**

* Either side of a floor division may now be rational.



v1.0.0-rc.8
====================

**Added:**

* `renf_elem_class::floordiv()` to perform `(a/b).floor()` more efficiently

**Changed:**

* changed renf_elem generator to always start with the zero element since it
  probably does not get generated by chance and can obviously cause a lot of
  trouble.



v1.0.0-rc.7
====================



v1.0.0-rc.6
====================



v1.0.0-rc.5
====================

**Added:**

* floor division of `renf_elem_t`
* added the missing implementations of free `floor`, `ceil`, and `pow` functions to the C++ library.

**Changed:**

* simplified testing of binary operators
* Comparison operators of `renf_elem_class` are now free so they participate in
  implicit casts in the same way like the boost generated free operators that
  are derived from them.

**Fixed:**

* Adapted to changes in unique-factory.
* implemented `renf_elem_class(renf_class, std::vector)` constructors that went missing in 39e1215e421e128b230fb62ff88ae0ac9b6414e7.

**Performance:**

* Speed up `renf_class::operator==`



v1.0.0-rc.3
====================

**Fixed:**

* release tarball was missing some header files with long names



v1.0.0-rc.2
====================

**Added:**

* ` renf_elem_swap(renf_elem_t, renf_elem_t)`
* There is now `e-antic/renfxx_fwd.h` if you only need forward declarations of
  `renf_class` and `renf_elem_class`.
* `renf_elem_class` can now be created from signed and unsigned long long.
* `renf_elem_class` can now be created from vectors of primitive integers, e.g.,
  ```
  renf_elem_class x(K, {1, 2, 3}); // = 3*x^2 + 2*x + 1
  ```
  where before the entries of the vector had to be `mpz_class`.
* Move semantics `&&` have been added to `renf_elem_class`.
* There is now support for serialization with cereal. See t-cereal.cpp for
  examples on how to use it.
* the release process has been automated with [rever](https://regro.github.io/rever-docs/)

**Changed:**

* `e-antic/renfxx.h` now requires C++17.
* All classes are now declared in the namespace `eantic`.
* The semantics of `operator =` have changed. In e-antic 0.1 the following
  would create the unit in the field K.
  ```
  renf_elem_class x(K);
  x = 1;
  ```
  Now the above statement makes x a rational number. More generally, an
  assignment resets the number field so that after `x = y` the condition
  `x.nf == renf_elem_class(y).nf` holds. To mimic the old behavior you need
  to rewrite the above as
  ```
  renf_elem_class x(K);
  x = renf_elem_class(K, 1);
  ```
* `renf_class` is now hidden behind a factory to get shared_ptr semantics
  everywhere. Create a `renf_class` by calling `renf_class::make(…)`. This
	returns a smart pointer, so you might have to replace some `.` with `->`.
* The change of semantics in assignment also affects reading from streams (in
  order to create `renf_elem_class`). Before the following would parse an element
  into a number field:
  ```
  renf_elem_class x(K);
  in >> x;
  ```
  Now this only works if the stream contains a rational number. (Otherwise an
  exception is raised.) As `in >> x` also resets `x.nf`. The above code should
  be replaced with:
  ```
  renf_elem_class x;
  K.set_pword(in);
  in >> x;
  ```
* `string renf_class::gen_name` is now a method so it needs to be called.
* Many operations that threw an exception before when domains were mixed, now
  abort program execution (typically through a call to `assert()`.) You are not
  supposed to mix domains unless explicitly stated otherwise.
* `renf_class.operator==` now also checks that the generator name is the same.
  Similarly, `renf_class.operator=` now also resets the generator name.

**Deprecated:**

* Some methods have been deprecated and might be removed in a future release,
  mostly to make the interface more consistent. The deprecation warnings give
  hints which methods to use instead.

**Removed:**

* `renf_elem_class(string&)` has been removed. If you want to parse a rational
  number, use `renf_elem_class(mpq_class(string))`. If you want to parse into a
  number field, use `renf_elem_class(renf_class&, string&)`.
* `renf_elem_class::operator=(string&)` has been removed. If you want to parse
  a rational into an element, use `x = mpq_class(string)`. If you want to parse
  into a number field, use `x = renf_elem_class(x.parent(), string)`.
* `renf_elem_class(vector<...>)` have been removed as it would have thrown an
  exception always anyway.
* `renf_elem_class::operator=(vector<...>)` have been removed due to the change
  of semantics of `=`. If `x` is not a rational you get the same behaviour as
  before with `x = renf_elem_class(x.parent(), {1, 2, 3})`.
* `renf_class::xalloc()` has been removed and replaced by an implementation
  detail.

**Fixed:**

* e-antic had claimed to be thread safe via an open MP pragma (in the number
  field refinement). In some cases, there was a problem with thread-safety. We
  now require users to explicitly mark multithreaded sections by forbidding
  mutations to a renf, see `renf_set_immutable`. As a result, there are some
  operations that cannot be done anymore in a multi-threaded environment but
  they now fail properly (instead of leading to random crashes.)



v0.1
====

**Added:**

* real roots isolation and refinement (poly_extra.h)
* renf structure (renf.h)
* renf_elem structures (renf_elem.h)
* C++ interface (renfxx.h)