path: root/doc/README.dev

Copyright 2002-2020 Free Software Foundation, Inc.
Contributed by the AriC and Caramba projects, INRIA.

This file is part of the GNU MPFR Library.

The GNU MPFR Library is free software; you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as published by
the Free Software Foundation; either version 3 of the License, or (at your
option) any later version.

The GNU MPFR Library is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
License for more details.

You should have received a copy of the GNU Lesser General Public License
along with the GNU MPFR Library; see the file COPYING.LESSER.  If not, see
https://www.gnu.org/licenses/ or write to the Free Software Foundation, Inc.,
51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.


Notes for the MPFR developers and Subversion users
==================================================

To compile source code obtained from the Subversion repository, you
need some GNU development utilities: aclocal, autoheader, automake,
autoconf 2.60 (at least), libtoolize, and the AX_PTHREAD macro from
<https://www.gnu.org/software/autoconf-archive/ax_pthread.html>
(for the latter, under Debian/Ubuntu it suffices to install the
autoconf-archive package).
As some files like "configure" are not part of the Subversion
repository, you first need to run "autoreconf -i" (or ./autogen.sh,
which could be used later to update the config files). Then you can
run ./configure in the usual way (see the INSTALL file, but note that
there are no patches to apply, and the URLs are not valid since the
corresponding version has not been released yet).

To generate mpfr.info, you need texinfo version 4.2 (or higher).

===========================================================================

The VERSION file contains the number of the next release version, i.e.
the version currently being developed. A suffix can be attached for the
development versions (in general, "-dev") or pre-release versions (e.g.
"-rc1"). It must be updated with the update-version script. Examples:

  tools/update-version 3 1 0 dev
  tools/update-version 3 1 0 rc1
  tools/update-version 3 1 0

The "-dev" suffix means that additional tests may be done or required
for development. For instance, the data files from the tests/data
directory are required; such files are not included in tarballs as
they can be large.

If nightly snapshots are built, the date in the yyyymmdd format and/or
the Subversion revision number (giving more accurate information) must
be added to the version as a suffix, for instance: "2.3.0-20070621" or
"2.3.0-r4553".

Patches can be tracked by adding a chunk of the form

--- PATCHES~    Tue Nov  6 19:59:33 2001
+++ PATCHES     Tue Nov  6 19:59:42 2001
@@ -1,0 +1 @@
+<your-id-here>

to the patch file[*]. After such patches have been applied, the file
src/get_patches.c providing the mpfr_get_patches() function will be
rebuilt by "make". MPFR distributors can still modify the version
suffix from the applied patches according to their version naming
scheme; for instance, for their own patches, MPFR developers do:

  tools/update-version 3 1 0 p1 -

[*] This idea comes from Thomas Roessler, who implemented it in Mutt.

For patches from MPFR developers, e.g. for MPFR 3.1.0:
1. Unarchive the tarball: a directory mpfr-3.1.0 is created.
2. Go into this directory (cd mpfr-3.1.0).
3. Apply the current patches with "patch -N -Z -p1 < /path/to/allpatches".
4. Reset the PATCHES file with "true >| PATCHES".
5. Rename mpfr-3.1.0 as mpfr-3.1.0-a and duplicate it as mpfr-3.1.0-b
   without changing the timestamps (e.g. with cp -a).
6. In mpfr-3.1.0-b, apply the patch obtained with "svn diff", e.g.
     patch --no-backup-if-mismatch -p0 < /path/to/new_patch
   If an autotools file has been modified, run "autoreconf" and remove
   the autom4te.cache directory.
7. In mpfr-3.1.0-b, update the version information:
     tools/update-version 3 1 0 p<n> -
   where <n> is the patch number.
8. In mpfr-3.1.0-b, update PATCHES file: echo >> PATCHES <patch_name>
9. Make the patch: TZ=UTC0 diff -Naurd mpfr-3.1.0-a mpfr-3.1.0-b

The tools/build-patch script can be used to ease the process.

Note: if autotools files are modified, the corresponding changes in the
distributed files depending on them must be included in the patch, and
the timestamps of such autotools files should be reset so that they do
not change when the patch is applied with the -Z option. Otherwise the
autotools would be needed to build MPFR (unless maintainer mode is
disabled).

Patches are put under the misc/www directory of the Subversion repository.
The web server is updated and patches are also copied as InriaForge files.

===========================================================================

When submitting patches, unified diffs (option -u) are recommended,
as they are more readable. You can also use the option -d to generate
a smaller set of changes. See diff(1) for more information.

===========================================================================

Copyright Notices: For easier maintainability, make sure that the
copyright notices match the regexp "Copyright.* yyyy Free Software"
where yyyy is the year of the latest modification in the branch
(and nothing else should match it).

The latest rules for GNU software can be found here:

  https://www.gnu.org/prep/maintain/maintain.html#Copyright-Notices

===========================================================================

To make a release (for the MPFR team):

  *** Please read this section entirely before making any release. ***

  Note: The following needs to be done in a branch x.y for MPFR x.y.z.

  0) Make sure that the src/mpfr-longlong.h file (from GMP's longlong.h)
     and the libtool-related files (config.guess, etc.) are up-to-date.
     Running autogen.sh may be necessary; avoid the possible warnings,
     as long as this does not require too recent tools (e.g. < 4 years
     old).

  1) Check the version and change the suffix to "rc1", "rc2", etc. with
     tools/update-version for the release candidates; remove the suffix
     for the final release.
     Update the libtool version (see src/Makefile.am).
     Update the DLL version (see configure.ac) if need be.
     Check these versions with tools/ck-version-info (this check will also
     be done automatically by "make dist" / "make distcheck").
     Update the date in doc/mpfr.texi.

  2) Generate the tuning parameters on different architectures and
     put them in src/mparam_h.in. For each architecture:

     a) download the latest release of GMP on gmplib.org
     b) build GMP with --disable-shared in say /tmp/gmp-x.y.z
        There is no need in tuning GMP, since most users will build MPFR
        with a vanilla GMP installation, i.e., with the default GMP tuning;
        however you need to go into /tmp/gmp-x.y.z/tune and type "make speed"
        (the MPFR tuning is using the resulting speed library)
     c) configure MPFR with --disable-shared --with-gmp-build=/tmp/gmp-x.y.z
     d) go into the "tune" directory and run "make tune"
     e) put the resulting mparam.h file into mparam_h.in (please include
        the version of GMP and the compiler used)

     You can produce time graphs to check the thresholds are correct (and
     compare to the corresponding mpf functions) with mbench. For example
     (-x1 corresponds to add, -x2 to sub, -x3 to mul, ...):

     $ cd mpfr/tools/mbench
     $ make mpfr-gfx GMP=... MPFR=...
     $  ./mpfr-gfx -b16 -e320 -s16 -f2 -x3 # compares mpfr_mul and mpf_mul
                                           # from 16 to 320 bits with increment
                                           # of 16 bits
     $ gnuplot -persist plot.gnuplot

     Another example, comparing mpfr_mul and mpf_mul from 2 to 1000000 bits,
     with ratio 1.1 between two sizes, 10 random values, and 10 smoothness
     checks:

     $ ./mpfr-gfx -b2 -e1000000 -r1.1 -f10 -x3 -m10
     $ gnuplot -persist plot.gnuplot

     Check the coverage of each source file by the test suite is at least 90%
     (or clearly justify any value under this threshold), and publish (for
     example in NEWS) the global coverage of this release. The individual
     coverage of each source file might also be published on the release web
     page. There is a specific mparam.h file to improve coverage; it should
     be tested by configuring MPFR with -DMPFR_TUNE_COVERAGE.

     Also test with -DMPFR_COV_CHECK, which allows one to check the coverage
     of some combinations of variable values (as defined in the MPFR source
     and test suite).

  3) Update the NEWS file, in particular say if the release is binary
     and/or API compatible (or not) with previous releases.
     Also update the "API Compatibility" section in the manual (mpfr.texi).
     Check with abi-compliance-checker (ABI Compliance Checker)[*], on the
     latest MPFR releases built with no configure options (except --prefix),
     that no changes have been missed. The ^/misc/build-multi script in the
     repository may be useful to prepare data for abi-compliance-checker.
     Note that abi-compliance-checker can only check the symbols, types
     and constants; it cannot detect just a change in the behavior, thus
     may miss some incompatibilities.
     Update the FAQ.html file with update-faq (and check it) in the doc
     directory.
     [*] https://lvc.github.io/abi-compliance-checker/

  4) Update the ChangeLog file with "TZ=UTC0 svn log -rHEAD:0 -v" in
     UTF-8 locales, e.g. "LC_ALL=en_US.UTF8 TZ=UTC0 svn log -rHEAD:0 -v".
     Make sure that all the pending commits have been done.

  5) Do a "svn export" of the branch to make sure to start from a clean
     source tree; this also has the advantage (over a checkout) to set
     the timestamps to the commit time, ensuring the right ordering of
     the files by date.
     Generate the tarballs with:
       $ ./autogen.sh
       $ ./configure
       $ make distcheck

  6) Test the release version on different machines, with --enable-assert
     set to "yes", "no" (default), "none" and "full" respectively, with
     and without -DMPFR_DISABLE_IEEE_FLOATS in $CFLAGS, with and without gmp
     internal files (--enable-gmp-internals), with and without GMP built as
     a shared library, with objdir equal to and different from srcdir (e.g.
     ../mpfr-source/configure after making mpfr-source read-only), with
     and without --enable-logging.

     Try different temporary allocation methods: GMP's --disable-alloca
     configure option (or compile GMP with --enable-alloca=debug and MPFR
     with --with-gmp-build to be able to get the memory leak errors); and
     -DMPFR_ALLOCA_MAX=0.

     Try different gcc versions with different options: with and without
     "-std=c99 -O3 -D_XOPEN_SOURCE=500", with and without "-ansi" (which
     allows to turn off features that are incompatible with ISO C90),
     with and without [-ansi] -pedantic-errors (which has the effect to
     disable extensions, such as long long when used together with -ansi),
     with and without -std=c11, with and without --enable-thread-safe, in
     various FPU precisions (double, double extended and single) if the
     platform supports that (e.g. under Linux/x86, with GCC and its -mpc64
     option to simulate the FreeBSD / NetBSD 6- behavior, where by default,
     the x87 FPU is configured to round on 53 bits), and in various locales
     (LC_ALL=tr_TR in particular, if installed).
     On x86, test with -m96bit-long-double and -m128bit-long-double.
     Try also with gcc's -fno-common option.
     Check also with "-Wformat=2", but without logging support (in order
     to avoid too many spurious warnings).
     Check with "-UHAVE_BIG_ENDIAN -UHAVE_LITTLE_ENDIAN" to simulate
     platforms where the endianness is unknown (or can't be specified
     without AC_CONFIG_HEADERS).
     Check also without the mpz_t pool (-DMPFR_POOL_NENTRIES=0).
     Check the generic code, e.g. with -DMPFR_GENERIC_ABI in $CFLAGS
     (useful because most tests are written for low precision) and with
     mpfr_cv_c_long_double_format=unknown (as a variable assignment).
     Check with -DMPFR_GROUP_STATIC_SIZE=0 to detect memory leaks that would
     occur in medium precision or more if a MPFR_GROUP_CLEAR were forgotten.

     Check that make and make check pass with a C++ compiler, for example:
     ./configure CC=g++ (MPFR 2.3.2 did not).
     Also test --enable-gmp-internals with it.

     Try different compilers, e.g., icc, opencc (x86_64 machines),
     tcc <https://bellard.org/tcc/>, llvm-gcc, clang.

     On 64-bit PowerPC, test against GMP built with the different ABI's:
     32, mode32 and mode64 (in particular mode32, where long's have
     32 bits and limbs have 64 bits [long long]).

     Test with -DMPFR_TESTS_FPE_DIV -DMPFR_ERRDIVZERO
     -DMPFR_DISABLE_IEEE_FLOATS in order
     to detect whether tests can fail due to a FP division by 0 (yielding
     either FE_DIVBYZERO, e.g. from 1.0 / 0.0 to generate an infinity, or
     FE_INVALID, e.g. from 0.0 / 0.0 to generate a NaN) on platforms where
     such an operation fails (e.g. trap). On platforms that do not support
     IEEE 754, such an operation yields an undefined behavior.
     If _MPFR_IEEE_FLOATS is defined to 1 (by the configure script), some
     divisions by 0 are avoided in the MPFR library.
     The -DMPFR_DISABLE_IEEE_FLOATS option sets _MPFR_IEEE_FLOATS to 0,
     allowing one to detect more issues, for platforms without IEEE floats.

     Test with -D_MPFR_PREC_FORMAT=2 when the "int" type is smaller
     than the "long" type.

     Test with mini-gmp.

     Test with valgrind by setting the environment variable:
       LOG_COMPILER="valgrind -q --error-exitcode=1 --leak-check=full"
     See below for more information about valgrind.

     Test with "clang -fsanitize=undefined" (available as of Clang 3.3),
     e.g.: ./configure CC=clang CFLAGS='-fsanitize=undefined'
     The -fno-sanitize-recover option can give more visibility by making
     the corresponding tests fail (useful for automated tests). However
     clang unconditionally regards the floating-point division by zero
     as an error with "-fsanitize=undefined"; this is detected by a
     configure test, which sets MPFR_ERRDIVZERO to disable the tests
     involving a floating-point division by zero. Alternatively, on systems
     supporting IEC 60559 / IEEE 754 division by zero, one can also provide
     the -fno-sanitize=float-cast-overflow,float-divide-by-zero option
     *after* the -fsanitize=undefined one.

     GCC 4.9 also supports "-fsanitize=undefined", but it just gives
     diagnostic messages at run time, not a failure; GCC 5 supports
     -fno-sanitize-recover like clang.

     Test with GCC's AddressSanitizer (-fsanitize=address). Optimizations
     should not be enabled, otherwise they can make some errors disappear.
     One also needs to unset LD_PRELOAD or use -static-libasan to avoid
     failures. But the -static-libasan solution does not work with MPFR,
     as it yields the following error:
       Your application is linked against incompatible ASan runtimes.

     Test with i686-w64-mingw32 under Wine (see below).

     Test with both "make check" and the worst cases.

     Check various warnings, in particular for obsolescent features.
     With GCC: "-Wall -Wold-style-declaration -Wold-style-definition
     -Wmissing-parameter-type -Wmissing-prototypes -Wmissing-declarations
     -Wmissing-field-initializers". The -Wint-in-bool-context option
     could be added once available. These warnings can easily be checked
     in automatic tests by adding "-Werror -Wno-error=unused-function",
     but this needs:
       * GCC 4.9+
       * a patched autoconf:
     https://lists.gnu.org/archive/html/autoconf-patches/2014-01/msg00003.html

     Check whether some functions could be declared as pure, const, etc.
     with GCC, using -Wsuggest-attribute=... options.

     Check that there are no abnormal regressions in the timings (both for
     100, 1000, 10000 digits, https://www.mpfr.org/mpfr-current/timings.html,
     and for small precision, using the mbench program, see mpfr/mbench).

     Test the library interface compatibility by running the test suite
     compiled against an old library version and dynamically linked with
     the new library version: for instance, build the shared library of
     old and new MPFR versions with the same configure options, and from
     the build directory of the old version, do something like:
     (cd src/.libs && \
      ln -nsf ../../../mpfr-new/src/.libs/libmpfr.so.1.* libmpfr.so.1)
     then "make check".

     Also test with different environment variables set
     (GMP_CHECK_RANDOMIZE, MPFR_CHECK_LIBC_PRINTF, MPFR_CHECK_LARGEMEM,
     MPFR_SUSPICIOUS_OVERFLOW, MPFR_CHECK_LOCALES).

     Check there is no branch misprediction due to wrong MPFR_LIKELY or
     MPFR_UNLIKELY statements. For that test, configure with
     --enable-debug-prediction, run "timings-mpfr 100", and check that
     the output contains no WARNING.

     For various platforms and compilers, check that:
       * [make check-gmp-symbols]
         MPFR does not use GMP internal symbols (unless --with-gmp-build
         or --enable-gmp-internals has been used);
       * [make check-exported-symbols]
         MPFR does not define symbols with a GMP reserved prefix.
     But note that these rules are not really portable: they may do
     nothing or might even incorrectly fail on some platforms.

  7) For the release itself (not the release candidates), if no problems
     have been found, create a tag with:
       svn cp .../mpfr/branches/x.y .../mpfr/tags/x.y.z

  8) For the release itself (not the release candidates), update
     the version with the update-version script to indicate the
     next version (use the "dev" suffix).

  9) * For the release itself (not the release candidates):
     Create a web page for the MPFR release and add the documentation
     (for mpfr.html, use "makeinfo --html --no-split mpfr.texi" from
     the doc directory). Make sure that both the .dvi and .ps files
     have an a4 papersize (see technical information later about the
     MPFR manual).
     Upload the tarballs and the signatures to the MPFR web server
     (via a svn working copy) and to InriaForge.
     Prepare the files for the GNU FTP site with the gnu-sigdir script
     in the /misc directory and upload them.
     Update the mpfr-current symbolic link and the history page.
     Update the old current page to point to the new release; see
     examples for 3.0.1 (latest version of the branch) and 3.1.0
     (which is not the latest version of the branch).
     Run the tools/announce-text script to do some checking and get
     the announce text. Edit this text if need be.
     Announce the release in the mpfr-announce, mpfr, gmp-discuss, gcc
     and info-gnu[1] mailing-lists, and on InriaForge (MPFR News, and
     contact the InriaForge administrators[2] to have the announce
     published on the main page).
     In case of a new patchlevel release, add a link from the web page
     of the previous release.

     * For the release candidates: Announce the RC in the mpfr-announce,
     mpfr, gmp-discuss, gcc and platform-testers[3] mailing-lists.
     A minimal web page for the MPFR release can be created right now
     (see svn history such as [4] for examples), as the manual already
     contains the new URL's.

     [1] https://www.gnu.org/prep/maintain/html_node/Announcements.html
     [2] http://siteadmin.gforge.inria.fr/FAQ.html
     [3] See https://lists.gnu.org/mailman/listinfo/platform-testers and
     https://lists.gnu.org/archive/html/platform-testers/2011-09/msg00000.html
     [4] https://gforge.inria.fr/scm/viewvc.php/misc/www/mpfr-3.1.2/index.html?view=markup&revision=8472&root=mpfr

Note: Mail sent to the mpfr-announce list should also be sent to
the mpfr list, and the Reply-To should be set to the mpfr list.

For major or minor releases (but not patchlevels), a branch may be
created first to allow new features to be committed to the trunk.

To add tcc support with libtool 2.4.2 or below, do the following before
running "make distcheck":
$ patch m4/libtool.m4 libtool-tcc-wl.patch
$ autoreconf

And for libtool 2.4.3 to 2.4.6, the following is needed:
$ patch m4/libtool.m4 libtool-tcc-rpath.patch
$ autoreconf

===========================================================================

Here is a non-exhaustive list of macros used for building and checking MPFR.
Most of them are automatically set up by the configure script and its options.

List of macros used for building MPFR (also used for checking):

+ HAVE_CONFIG_H:        Define if we have to include 'config.h' first.
+ MPFR_HAVE_GMP_IMPL:   Define if we have the gmp internal files.
                        ('gmp-impl.h', 'gmp-maparam.h', ...).
+ MPFR_USE_MINI_GMP:    Define to use mini-gmp.

+ HAVE_ALLOCA_H:        Define if the function alloca() is in alloca.h.
+ HAVE_LOCALE_H:        Define if <locale.h> is available.
+ HAVE_LONG_LONG:       Define if the system supports 'long long'.

+ HAVE_STDARG:          Define if the system supports 'stdarg.h'.
                        Otherwise it is assumed it is 'vararg.h'.

+ HAVE_INTTYPES_H:      Define if <inttypes.h> is available (ISO C99).
+ HAVE_STDINT_H:        Define if <stdint.h> is available (ISO C99).
+ MPFR_HAVE_INTMAX_MAX: Define if the INTMAX_MAX macro works correctly
                        (if 'intmax_t' is supported).

+ MPFR_HAVE_C11_LOCK:   Define if C11 threads are supported.
+ HAVE_PTHREAD:         Define if pthread is available.

Format of long double.
+ HAVE_LDOUBLE_IS_DOUBLE:               IEEE double.
+ HAVE_LDOUBLE_IEEE_EXT_BIG:            IEEE extended, big endian.
+ HAVE_LDOUBLE_IEEE_EXT_LITTLE:         IEEE extended, little endian.
+ HAVE_LDOUBLE_IEEE_QUAD_BIG:           IEEE quad, big endian.
+ HAVE_LDOUBLE_IEEE_QUAD_LITTLE:        IEEE quad, little endian.
+ HAVE_LDOUBLE_MAYBE_DOUBLE_DOUBLE:     Double-double (a.k.a. IBM).

+ MPFR_DISABLE_IEEE_FLOATS:
                        Related to the native floating-point types (e.g.
                        for conversion functions), use the generic code
                        instead of IEEE 754 specific one.
                        Note: This is mainly for developers in order to
                        check the generic code, as machines without IEEE
                        floating-point types are very uncommon nowadays.
+ MPFR_WANT_ASSERT:     Assertion level. See src/mpfr-impl.h for details.
+ MPFR_EXP_CHECK:       Define if we want to check the exp field.

+ _MPFR_PREC_FORMAT:    Used to define the mpfr_prec_t type.
+ _MPFR_EXP_FORMAT:     Used to define the mpfr_exp_t type.
                        Note: these two macros are for internal use,
                        testing and experimented users only; they must
                        not be changed when the MPFR library is to be
                        installed in a system directory.

+ IEEE_DBL_MANT_DIG:    Number of bits in the significand (mantissa) of a
                        double (default: 53).
+ MPFR_LDBL_MANT_DIG:   Number of bits in the significand (mantissa) of a
                        long double (generally based on the standard macro
                        LDBL_MANT_DIG). Note: be careful with formats such
                        as double-double (a.k.a. IBM long double).

+ MPFR_USE_LOGGING:     Define to enable logging.
+ MPFR_WANT_DECIMAL_FLOATS:
                        Define to build conversion functions from/to
                        decimal floats.
+ MPFR_WANT_FLOAT128:   Define to build conversion functions from/to
                        binary128 floats (_Float128 or __float128).

+ MPFR_ALLOCA_MAX:      Maximum size for the use of alloca by temporary
                        allocations (default: 16384).
                        This macro is not used when MPFR is built with
                        the GMP build directory (--with-gmp-build).

+ MPFR_USE_THREAD_SAFE: Define to build MPFR as thread safe (TLS).
+ MPFR_USE_C11_THREAD_SAFE:
                        Define to implement TLS in the C11 way.

+ MPFR_WANT_SHARED_CACHE:
                        Define to have caches shared by all threads.
+ MPFR_THREAD_LOCK_METHOD:
                        When MPFR_WANT_SHARED_CACHE is defined, this macro
                        gives the thread locking method (string).

+ MPFR_HAVE_NORETURN:   Define if the _Noreturn function specifier is
                        supported.
+ MPFR_HAVE_BUILTIN_UNREACHABLE:
                        Define if the __builtin_unreachable GCC built-in is
                        supported.

+ MPFR_GENERIC_ABI:     Define to disable code that is tied to a specific
                        ABI (e.g. GMP_NUMB_BITS value).
                        Note: Currently it is also used to disable code
                        specific to low precision, i.e. to use only generic
                        code. This is useful because most tests are written
                        for low precision, meaning that without this macro,
                        the generic code would not sufficiently be tested.

+ MPFR_WANT_PROVEN_CODE:
                        Define to enable formally proven code.

List of macros used for checking MPFR:

+ MPFR_HAVE_FESETROUND: Define if the function fesetround() is available
                        (and in header <fenv.h>).
+ MPFR_FPU_PREC:        Allows to test MPFR on x86 processors when the
                        x87 FPU rounding precision has been changed (see
                        tests/tests.c for its usage).
+ HAVE_SUBNORM_DBL:     Define if the double type fully supports subnormals.
+ HAVE_SUBNORM_FLT:     Define if the float type fully supports subnormals.
+ HAVE_SIGNEDZ:         Define if signed zeros are supported.
+ HAVE_SYS_TIME_H:      Define if the header sys/time.h is usable.
+ HAVE_GETTIMEOFDAY:    Define if the function gettimeofday() is available.
+ HAVE_SETLOCALE:       Define if the function setlocale() is available.
+ MPFR_ERRDIVZERO:      Define if the floating-point division by 0 fails
                        in the MPFR library (e.g., because a SIGFPE signal
                        is generated, or because it is regarded as undefined
                        behavior by a sanitizer). This disables the tests
                        involving such operations.
                        Note: Tests related to NaN and infinities must not
                        rely on native FP division by 0, whether this macro
                        is defined or not.
+ MPFR_TESTS_FPE_DIV:   Define to check whether there has been a FP
                        exception FE_DIVBYZERO or FE_INVALID, which
                        probably comes from 1.0 / 0.0 or 0.0 / 0.0 to
                        generate an infinity or a NaN. This is normally
                        used together with MPFR_ERRDIVZERO, in order to
                        check that all divisions by 0 have been protected
                        in the tests (so that tests can pass on platforms
                        where the floating-point division by 0 fails).
+ MPFR_TESTS_FPE_TRAP:  Define to trap the FE_DIVBYZERO and FE_INVALID
                        exceptions; MPFR_TESTS_FPE_DIV needs to be defined
                        too, and MPFR_ERRDIVZERO should be defined as well
                        to avoid spurious traps (see above).
+ MPFR_TESTS_TIMEOUT:   Define to enable timeout in the tests. Its value
                        specifies the default timeout (in seconds), or 0
                        for no timeout by default. When defined, this
                        value can be overridden at run time (e.g., with
                        "make check" or when executing an individual test)
                        with the MPFR_TESTS_TIMEOUT environment variable.
+ MPFR_TESTS_ABORT:     Define to replace exit(1) by abort(), thus with a
                        core dump.
+ MPFR_COV_CHECK:       Define to enable value coverage checking (must not
                        be used in production). This macro is for the MPFR
                        developers, in order to improve the test suite.

===========================================================================

Environment variables that affect the tests:

+ GMP_CHECK_RANDOMIZE:  Seed for the random functions, except for 0 or 1,
                        in which case a random (time based) seed is used.
                        By default, a fixed seed is used. Only developers
                        and testers should change the seed.

+ MPFR_CHECK_LARGEMEM:  Define to enable expensive tests.

+ MPFR_CHECK_LIBC_PRINTF:
                        Define to enable comparisons with the printf
                        function of the C library. These comparisons are
                        disabled by default as failures could be due to
                        the C library itself on some machines, and they
                        do not affect MPFR.

+ MPFR_CHECK_LOCALES:   Fail in case a locale cannot be set. Developers
                        can set this variable on their machines to make
                        sure that needed locales are properly installed
                        and tested.

+ MPFR_DEBUG_BADCASES:  For debugging (see tests.c, function bad_cases).

+ MPFR_SUSPICIOUS_OVERFLOW:
                        Define to check suspicious overflow in the generic
                        tests (tgeneric.c). For developers and testers.

+ MPFR_TESTS_MEMORY_LIMIT:
                        The memory limit for the tests (default is
                        2^22 = 4 MB). Set to 0 for unlimited.

+ MPFR_TESTS_TIMEOUT:   When timeout in the tests is enabled, this
                        overrides the value of the macro.

===========================================================================

Before testing any macro in a .c file, one needs:

#ifdef HAVE_CONFIG_H
# include "config.h"
#endif

except if mpfr-impl.h (for the library) or mpfr-test.h (for the tests) is
included first, because these header files already have the above code.

===========================================================================

The GNU Coding standards can be read at:
  https://www.gnu.org/prep/standards/

ISO C Names and corresponding headers:
  http://www.schweikhardt.net/identifiers.html

The C language:
  http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1256.pdf (C99)
  http://www.open-std.org/jtc1/sc22/wg14/www/C99RationaleV5.10.pdf
  http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1570.pdf (C11 draft)
  http://home.datacomm.ch/t_wolf/tw/c/c9x_changes.html

About undefined behavior:
  https://blog.regehr.org/archives/213
  https://blog.regehr.org/archives/226
  https://blog.regehr.org/archives/232
  https://blog.regehr.org/archives/1520
  http://blog.llvm.org/2011/05/what-every-c-programmer-should-know.html

Type punning, strict aliasing, and optimization:
  https://blog.regehr.org/archives/959

To allow MPFR to be built on some buggy compiler, try to follow
these rules:

   =====================================================================

Don't write:
  mp_limb_t l;
  [...]
  if (l) do_action ();
But:
  mp_limb_t l;
  [...]
  if (l != 0) do_action ();

since mp_limb_t may be "unsigned long long", and some buggy compiler
produce illegal codes with the first form.

   =====================================================================

Try to avoid "LONG_MIN/1" since it produces a SIGNAL on (old) FreeBsd.
Don't forget that LONG_MIN/-1 is not representable (specially
with code like MPFR_EXP_MIN/n).

   =====================================================================

Don't use "near" and "far" as variable names since they are "Keywords"
for some C compiler (Old DOS compiler). Also don't use "pm", which is used
by the C compiler 'sharp' to design variables that should be stored in the
flash memory. Don't use "new", which is reserved in C++.

Check C++ reserved keywords, e.g. from

  https://en.cppreference.com/w/cpp/keyword

or more generally:

  https://www.google.com/search?q=%22C%2B%2B%22+reserved+keywords

Quoted from <https://www.gnu.org/software/gcc/codingconventions.html>:

  Avoid the use of identifiers or idioms that would prevent code
  compiling with a C++ compiler. Identifiers such as new or class,
  that are reserved words in C++, should not be used as variables
  or field names. Explicit casts should be used to convert between
  void* and other pointer types.

When a string literal ("...") is followed by a macro name, there
must be white space between them, otherwise this is parsed as a
user-defined string literal in C++11:

  https://en.cppreference.com/w/cpp/language/user_literal
  https://stackoverflow.com/a/6402166/3782797

In at least mpfr.h, use the underscore version of the attribute names
(e.g. "__sentinel__" instead of "sentinel"), otherwise user code could
fail to compile with GCC when it defines macros such as "sentinel"
(before the #include's or via the -D command-line option). See

  https://gcc.gnu.org/onlinedocs/gcc/Attribute-Syntax.html

   =====================================================================

Setting errno is safe to signal some error information (as in the
formatted output functions), but errno must not be read (unless we
have just modified it) as this may yield undefined behavior in some
corner cases out of our control (ISO C99 / C11, 7.14.1.1p5, also
mentioned in J.2).

   =====================================================================

C-Reduce may be useful to try to identify whether a bug comes from the
compiler.

   =====================================================================

About type conversions:

To do type punning (i.e. store a value of some type and reinterpret
it as another type), use a union. This is valid in ISO C99 and above
(in C99, see 6.5#7 and Note 82 of 6.5.2.3#3 for the clarification),
but not in C++. So, users of a C++ compilers should make sure that
their compiler supports type punning via a union. If some problem is
reported, we should address it either by making the code compatible
or by adding a configure test to reject the compiler.

Some references:
* https://en.wikipedia.org/wiki/Type_punning#Use_of_union
* https://stackoverflow.com/questions/346622/opinions-on-type-punning-in-c
  "Opinions on type-punning in C++?"

Do not do conversions between function pointers and other kinds of
pointers (including to void *). This yields undefined behavior and
may not work in practice. Example:

  https://stackoverflow.com/questions/5579835/c-function-pointer-casting-to-void-pointer

Adding a level of indirection is OK, as suggested there, and on:

  https://stackoverflow.com/questions/36645660/why-cant-i-cast-a-function-pointer-to-void

   =====================================================================

For floating-point constants, do not use the non-standard and useless
suffix "D". It seems to mean "double" for GCC[*], but for instance,
ICC 15 regards 1.0D as 0 (though ICC claims compatibility with GCC)
and for tcc 0.9.27, this is an error. The standard suffixes from TS 18661-2
are:

	f l F L df dd dl DF DD DL

Moreover, avoid native floating-point if possible. Be careful that GCC
does not conform to the C standard by default. References:
  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=323
  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85957

[*] https://stackoverflow.com/questions/4331200/what-do-f-and-d-mean-at-the-end-of-numeric-literals

===========================================================================

Avoid variable names "l", "I" and "O", which look like "1" and "0" with
some fonts.

===========================================================================

For identifiers defined in MPFR, do not use the GMP namespaces
(gmp_..., GMP_...).

===========================================================================

You are allowed to use the mpn and mpz classes of GMP functions (types
and functions starting with "mpn_" and "mpz_"). However, except for some
conversion functions where they may be needed,
  * the mpq class and GMP's formatted output and input functions (i.e.,
    printf and scanf style) can only be used in an alternative method
    by testing MPFR_USE_MINI_GMP (and only if there is a real benefit),
    since they are not available in mini-gmp;
  * the mpf class must not be used at all.

===========================================================================

The headers <limits.h>, <stdio.h>, <stdlib.h> and <string.h> are always
included in mpfr-impl.h; thus you need not (and should not) include them
in usual source and test files.

===========================================================================

For files that need intmax_t or similar, use:

#if HAVE_INTTYPES_H
# include <inttypes.h>
#endif
#if HAVE_STDINT_H
# include <stdint.h>
#endif

Note that even though the ISO C99 standard requires that <inttypes.h>
include <stdint.h>, in practice this is not true on all platforms,
such as OSF/1 (Tru64) 5.1. This is consistent with autoconf, which
has used this form since 2004-01-26 (in headers.m4).

References:
  https://git.savannah.gnu.org/gitweb/?p=autoconf.git;a=commitdiff;h=62ac9bbfebe879581dabeed78c6ac66b907dd51d
  https://sympa.inria.fr/sympa/arc/mpfr/2010-08/msg00015.html

===========================================================================

Use locale-dependent functions when the result needs to depend on the
locales, e.g. the decimal-point character in mpfr_out_str.

Conversely, do not use locale-dependent functions when the result must
not depend on the locales. In particular, the alphanumeric characters
used in number strings (as created by mpfr_get_str) must be those of
the required characters from the basic character set (see ISO C99
standard Section 5.2.1 "Character sets"). And tolower(letter) does
not necessarily return the corresponding lowercase letter from these
required characters. For instance, tolower('I') returns a dotless 'i'
in Turkish tr_TR.iso88599 locales.

===========================================================================

If you have to mix TMP_DECL and MPFR_SAVE_EXPO_DECL in the declaring
section of your function, please declare MPFR_SAVE_EXPO_DECL before
TMP_DECL, since TMP_DECL may be replace by nothing:

Instead of:                                 Usually preprocessed as:
  unsigned long t                             unsigned long t;
  TMP_DECL (marker);                          ;
  MPFR_SAVE_EXPO_DECL (expo);                 mpfr_save_expo_t expo;
use:
  unsigned long t                             unsigned long t;
  MPFR_SAVE_EXPO_DECL (expo);                 mpfr_save_expo_t expo;
  TMP_DECL (marker);                          ;

===========================================================================

Memory allocation
-----------------

Do not use TMP_DECL / TMP_ALLOC, ... but MPFR_TMP_DECL, MPFR_TMP_ALLOC, ...

In the tests, use only tests_allocate, tests_reallocate and tests_free
(there may be some rare exceptions, such as in tabort_defalloc*.c).

Avoid code that would yield unnecessary reallocations, which can be very
expensive. In particular, for code that is based on the mpz layer of GMP,
do not use mpz_init, but mpz_init2 with the estimated maximum size; it is
better to overestimate this size a bit than underestimating it.

===========================================================================

Do not use C99-only features, such as empty macro arguments or C++-style
comments.

===========================================================================

When testing a "boolean" macro M (i.e. which is normally either equal
to 1 or undefined), do not use #if M, but #ifdef M or #if defined(M).
With icc, the form #if M triggers a warning ("remark #193: zero used
for undefined preprocessing identifier").

===========================================================================

If you want to use the logging of MPFR, you need to enable it:
  make distclean
  ./configure --enable-logging
  make
Then link your program with this new build of MPFR.

Warning! The logging code for functions sometimes output an "inexact"
value, but in case of exception, this value may be meaningless. In
fact, the output value is the value of some variable; please check
the source code of the function to understand its real meaning.

You can control what is logged using the environment variables:

MPFR_LOG_FILE:  Name of the LOG file (default: mpfr.log).
MPFR_LOG_FLUSH: When this variable is set, flush the log stream after
                each log output (useful to get the latest logs in case
                of crash, but this makes logging slower).
MPFR_LOG_PREC:  Number of digits of the output (set the internal variable
                mpfr_log_prec, default: 6).
MPFR_LOG_LEVEL: Max recursive level (default: 7).

MPFR_LOG_INPUT:    Log the input
MPFR_LOG_OUTPUT:   Log the output
MPFR_LOG_TIME:     Log the time spent inside the function.
MPFR_LOG_INTERNAL: Log the intermediary variables if any.
MPFR_LOG_MSG:      Log the messages sent by MPFR if any.
MPFR_LOG_ZIV:      Log what the Ziv Loops do.
MPFR_LOG_STAT:     Log how many times Ziv failed.
MPFR_LOG_ALL:      Log everything

Define them. Run your program, and view `mpfr.log`.

For example, just define MPFR_LOG_ALL, run you program, and view `mpfr.log`.

Note: The running time may be much longer. If logging is used on the
test suite with a default timeout, it may be necessary to increase the
timeout time by setting the environment variable MPFR_TESTS_TIMEOUT
to the new timeout value in seconds (or 0 to disable the timeout).

===========================================================================

This feature is available only for gcc >= 3.0 and glibc >= 2.0.
To achieve this, these macros have been added:

+++ MPFR_LOG_VAR(y)
 Log a MPFR variable if requested (INTERNAL).
 Example:
   mpfr_t y;
   MPFR_LOG_VAR (y);

+++ MPFR_LOG_MSG(x)
 Log another message (a warning for example)
Example:
 MPFR_LOG_MSG (("WARNING: Unchecked code\n", 0));
The 0 is here a dummy value, because there must be at least an argument
after the format string.

+++ MPFR_LOG_BEGIN(x)
 Add this macro at the beginning of a function.
Example:
 int dodo (mpfr_t x, mpfr_t op, int cnt, mpfr_rnd_t rnd) {
  [decl]
  MPFR_LOG_BEGIN (("op[%Pu]=%.*Rg rnd=%s",
                    mpfr_get_prec(op), mpfr_log_prec, op, RND2STR(rnd)));

+++ MPFR_LOG_END(x)
 Add this macro at the end of a function.
Example:
 MPFR_LOG_END (("x[%Pu]=%.*Rg i=%d", mpfr_get_prec (x), mpfr_log_prec, x, i));
 return i;
}

+++ MPFR_LOG_FUNC (begin,end)
  Add this macro at the beginning of a function. It does
the same job as MPFR_LOG_BEGIN and MPFR_LOG_END but it is smatter
since it intercepts the return itself to put the end statement.
Example
 MPFR_LOG_FUNC (
    ("op[%Pu]=%.*Rg rnd=%d", op, mpfr_get_prec (op), mpfr_log_prec, op),
    ("x[%Pu]=%.*Rg inexact=%d", mpfr_get_prec (x), mpfr_log_prec, x, i));


The double brackets "((" and "))" are needed since MPFR must still
compile with non GNU compiler, so Macros with variable # of args
are not allowed.

It uses the extension of the mpfr_printf function: %Rf to display a mpfr_t.
%Ru is used to display the precision of a mpfr_t.
It uses some extended attributes of GCC (constructor, etc.) to achieve
its goals too.

===========================================================================

ZivLoop Controller

Ziv strategy is quite used in MPFR. In order to factorize the code, you
could use these macros:

+++ MPFR_ZIV_DECL(_x)
 Declare a ZivLoop controller

+++ MPFR_ZIV_INIT(_x, _prec)
 Init a ZivLoop controller according to the initial value of _prec.

+++ MPFR_ZIV_NEXT(_x, _prec)
 Increase the precision _prec according to the ZivLoop controller.

+++ MPFR_ZIV_FREE(_x)
 Free the ZivLoop controller.

===========================================================================

If you plan to add a new function, you could follow this schema:

int
mpfr_toto (mpfr_ptr rop, mpfr_srcptr op, mpfr_rnd_t rnd)
{
  [Declare all used variables]
  int inexact;
  mpfr_prec_t prec;
  MPFR_ZIV_DECL (loop);
  MPFR_SAVE_EXPO_DECL (expo);

  /* Log it if requested */
  MPFR_LOG_BEGIN
    (("op[%Pu]=%.*Rg rnd=%d", mpfr_get_prec (op), mpfr_log_prec, op, rnd),
     ("rop[%Pu]=%.*Rg inexact=%d",
       mpfr_get_prec (rop), mpfr_log_prec, rop, inexact));

  /* First deal with particular cases */
  if (MPFR_UNLIKELY (MPFR_IS_SINGULAR (op)))
    {
      if (MPFR_IS_NAN (op))
        {
         MPFR_SET_NAN (rop);
         MPFR_RET_NAN;
        }
      else if (MPFR_IS_INF (op))
        {
         [Code to deal with Infinity]
        }
      else
        {
          MPFR_ASSERTD (MPFR_IS_ZERO (op));
          [Code to deal with Zero]
        }
    }
  [Other particular case: For example, op<0 or op == 1]

  [Compute the first estimation of the used precision `prec`]
  [Initialize the intermediate variables using mpfr_init2]
  MPFR_SAVE_EXPO_MARK (expo);            /* Maximal range for exponent */

  MPFR_ZIV_INIT (loop, prec);            /* Initialize the ZivLoop controller */
  for (;;)                               /* Infinite loop */
    {
      [Compute an estimation of the function and]
      [an estimate of the error.]
      if (MPFR_CAN_ROUND (...))          /* If we can round, quit the loop */
        break;
      MPFR_ZIV_NEXT (loop, prec);        /* Increase used precision */
      [Use `mpfr_set_prec` to resize all needed intermediate variables]
    }
  MPFR_ZIV_FREE (loop);                  /* Free the ZivLoop Controller */

  inexact = mpfr_set (rop, temp, rnd);   /* Set rop to the computed value */
  [Clear all intermediate variables]

  MPFR_SAVE_EXPO_FREE (expo);            /* Restore exponent range */
  return mpfr_check_range (rop, inexact, rnd);  /* Check range and quit */
}

Make sure that Ziv loops cannot increase the precision forever because of
internal exception. Otherwise one gets either a segmentation fault (with
limited stack size) or an assertion failure (with unlimited stack size,
e.g. with "make check").

Do not use code with side effects inside MPFR_ASSERTD or MPFR_ASSERTN,
as assertion checking can be disabled. If a variable is set only to test
it in an MPFR_ASSERTD expression, the MPFR_DBGRES macro should be used
with the assignment as its argument, e.g.
  int inex;
  MPFR_DBGRES (inex = mpfr_set (y, x, rnd));
  MPFR_ASSERTD (inex == 0);

Exception handling (overflow/underflow in particular):
  * Warning: To detect exceptions and/or possible error loss due to
    internal exceptions, testing whether some variable is singular with
    MPFR_IS_SINGULAR is generally not sufficient! Indeed, in case of
    overflow (resp. underflow), the value may be rounded (in absolute
    value) to the largest finite number (resp. to the smallest non-zero
    number, possible even in round-to-nearest mode).
  * The MPFR_BLOCK* macros can be useful, e.g.
      {
        MPFR_BLOCK_DECL (flags);
        /* ... */
        MPFR_BLOCK (flags, /* expression or statements */)
        /* ... */
        if (MPFR_OVERFLOW (flags))
          { /* case of overflow in expression or statements */ }
        /* ... */
      }
    See mpfr-impl.h (search for MPFR_BLOCK) for more information.

===========================================================================

If you plan to add a new threshold in MPFR which could be tuned,
you should add its default value in the file `mparam_h.in'. When the
script configure finishes, it creates the file `mparam.h' from `mparam_h.in'.

Then you needs to modify the program `tuneup.c' to allow it to compute
the new threshold. If it is a classical threshold (not complex), you could
use this method (example of mpfr_exp):

/* Define the threshold as a variable instead of a constant */
mpfr_prec_t mpfr_exp_threshold;
#undef  MPFR_EXP_THRESHOLD
#define MPFR_EXP_THRESHOLD mpfr_exp_threshold
/* Include the test function to threshold directly in the test
   program. It will override the mpfr_exp coming from libmpfr.a */
#include "exp.c"
/* Define the speed function related to mpfr_exp */
static double speed_mpfr_exp (struct speed_params *s) {
  SPEED_MPFR_FUNC (mpfr_exp);
}

Then in the function `all', you will have to call the tune function,
and write the new THRESHOLD in the file `mparam.h':

  /* Tune mpfr_exp */
  if (verbose)
    printf ("Tuning mpfr_exp...\n");
  tune_simple_func (&mpfr_exp_threshold, speed_mpfr_exp);
  fprintf (f, "#define MPFR_EXP_THRESHOLD %lu\n",
           (unsigned long) mpfr_exp_threshold);

More complex tuning is possible but needs special attention.

===========================================================================

MPFR uses many macros, thus finding where an error occurs exactly may
be difficult when it is in some macro expansion. For GCC users, a new
experimental -ftrack-macro-expansion option has been added in GCC 4.7.
"It allows the compiler to emit diagnostic about the current macro
expansion stack when a compilation error occurs in a macro expansion."
<https://gcc.gnu.org/gcc-4.7/changes.html>

===========================================================================

Bit Twiddling Hacks - Sean Eron Anderson maintain a list of tricks to get
efficient code on <https://graphics.stanford.edu/~seander/bithacks.html>.
WARNING: some of those tricks may not take into account possible overflows,
and may not be portable.

===========================================================================

MPFR manual (mpfr.texi):
  * Use "significand", not "mantissa".
  * Use "decimal-point character" (as in ISO C) when the meaning is
    the corresponding character ("." or ",", depending on the locale),
    not "decimal point", "decimal separator", "fractional point", or
    "radix point" (the latter is fine in certain contexts). This is
    just the name of this character (as initially defined for base 10)
    and does not imply a number written in decimal.
  * Use "@minus{}" for the minus character, not "-".
  * Warning! Texinfo is not like TeX. Whitespace is preserved in the
    info file. So, do not use additional space for .texi indentation.
    This also means that you need to care about the typography. Please
    read Section "Inserting Space" in the Texinfo manual.
  * Follow the English typography (no space before punctuation marks,
    double space after a period, etc.), not the French one.

The MPFR manual in DVI/PS/PDF formats should have an a4 papersize, as
declared in mpfr.texi (@afourpaper command). The DVI file format (.dvi)
traditionally does not contain papersize information, which has two
consequences:
  * When viewing a .dvi file, one gets the papersize from global settings,
    which may differ from the papersize declared in the mpfr.texi file.
    In particular, a4 text on letter paper can be truncated, depending on
    the margins.
  * Since the .ps file is built from the .dvi file and makeinfo does not
    provide the papersize information to the dvips command, the .ps file
    can get a wrong papersize, depending on the settings on the machine
    where this file is generated. Papersize information should be checked
    before distributing the .ps file.
Nowadays, .dvi files can contain papersize information via "specials",
and texinfo.tex has been updated to include such information. However,
the interpretation of such data is based on a common agreement between
drivers rather on a standard. In short, the papersize issues should no
longer appear, but this should be checked manually. The bug report:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=874632

===========================================================================

Running "make" outputs a lot of information, and warnings are not very
visible. The following tool "eet" allows a copy of warning messages to
be output to a different window (e.g. xterm or zenity):

  https://www.vinc17.net/unix/#eet

Direct link to the tarball: https://www.vinc17.net/unix/eet.tar.xz

===========================================================================

Be careful when avoiding "'var' may be used uninitialized in this function"
warnings from gcc. Initializing such variables to a dummy value has several
drawbacks:
  * this may prevent other tools (that do static or dynamic analysis) from
    detecting bugs;
  * this makes code maintenance more difficult (e.g. when modifying the
    code, one may more easily forget a real initialization);
  * this makes the compiler add useless code (though this should not be
    significant).

The INITIALIZED macro can be used to avoid such warnings with gcc, e.g.

  int INITIALIZED(i);

It uses the "int i = i;" pseudo-initialization trick, disabled with other
compilers as this is undefined behavior. See:

  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36296

If a dummy initialization must be added, use preferably an "invalid" value
(e.g. NULL for pointers, or a value that can be checked with MPFR_ASSERTN
before using it) that could make the program abort instead of returning an
incorrect value in case of a bug in MPFR.

===========================================================================

Avoid mixing signed and unsigned integer types, as this can lead signed
types to be automatically converted into unsigned types (usual arithmetic
conversions). If such a signed type contains a negative value, the result
may be incorrect on some platforms. With MPFR 2.x, this problem could
arise with mpfr_exp_t, which is signed, and mpfr_prec_t (mp_prec_t),
which was unsigned (it is now signed), meaning that in general, a cast
of a mpfr_prec_t to a mpfr_exp_t was needed.

Note that such bugs are difficult to detect because they may depend on
the platform (e.g., on LP64, 32-bit unsigned int + 64-bit long will give
a signed type, but on ILP32, 32-bit int + 32-bit unsigned long will give
an unsigned type, which may not be what is expected), but also on the
input values. So, do not rely on tests very much. However, if a test
works on 32 bits but fails on 64 bits in the extended exponent range
(or conversely), the cause may be related to the integer types (e.g. a
signness problem or an integer overflow due to different type sizes).

For instance, in MPFR, such issues were fixed in r1992 and r5588.

An example that will fail with 32-bit int and long:

    long long umd(void)
    {
      long a = 1;
      unsigned int b = 2;
      return a - b;
    }

When creating a new variable that will always contain nonnegative values,
it is generally better to define it as a signed type if it may be used in
an arithmetic expression. The exceptions are when the value is seen as an
array of bits (e.g. for limbs) and to temporarily avoid integer overflow.

===========================================================================

To use features related to types larger than type long, "mpfr-intmax.h"
must be included before "mpfr-impl.h".

The intmax_t and uintmax_t types can be used only if _MPFR_H_HAVE_INTMAX_T
is defined. In this case, the printf / gmp_printf length specifier "j" can
be used only when NPRINTF_J is not defined.

For internal use, mpfr-intmax.h also unconditionally defines mpfr_intmax_t,
mpfr_uintmax_t, MPFR_UINTMAX_MAX, MPFR_INTMAX_MAX, MPFR_INTMAX_MIN and
the corresponding length specifier MPFR_INTMAX_FSPEC.

Warning! mpfr_intmax_t may be smaller than intmax_t if NPRINTF_J is defined.

===========================================================================

Use mpfr_prec_t and mpfr_rnd_t instead of the old types mp_prec_t and
mp_rnd_t. Similarly, use mpfr_exp_t instead of GMP's mp_exp_t type
(unless you really want mp_exp_t, e.g. for conversions with mpf; but
you must not assume that mpfr_exp_t and mp_exp_t are identical).

===========================================================================

How to specify (for reading) the minimum exponent or the maximum exponent
in the MPFR source depends on the context.

1. The most portable form is mpfr_get_emin() and mpfr_get_emax(). In
   the MPFR source, this is equivalent to __gmpfr_emin and __gmpfr_emax
   respectively (macros are defined in mpfr-impl.h; the only difference
   is that the macros do not evaluate to a lvalue).

2. If the exponent range has been extended, you can use the constants
   MPFR_EMIN_MIN and MPFR_EMAX_MAX instead. This will be faster if TLS
   is enabled. It also avoids a bug on some Linux/Sparc machines with
   some GCC versions and TLS, but this shouldn't be the primary concern,
   as this might be the other way round on some other machines. This is
   the most common context.

3. If you want the minimum and maximum possible exponent values supported
   by MPFR, use MPFR_EMIN_MIN and MPFR_EMAX_MAX respectively.

4. If you want the minimum and maximum values supported by the mpfr_exp_t
   type (i.e. the limits of this type), use MPFR_EXP_MIN and MPFR_EXP_MAX
   respectively. This may be useful for intermediate computations on the
   exponents.

More on exponent handling:

* The mpfr_exp_t type has at least 32 bits since it must contain the
  default exponent range.

* The range of valid exponents is defined so that if a and b are two
  valid exponents (i.e. between MPFR_EMIN_MIN and MPFR_EMAX_MAX), then
  ± a ± b ± 1 fits in a mpfr_exp_t.

* The unsigned type corresponding to mpfr_exp_t is mpfr_uexp_t. It may be
  useful if the considered values are nonnegative and don't necessarily
  fit in mpfr_exp_t. To convert a nonnegative mpfr_exp_t to mpfr_uexp_t,
  you should use the MPFR_UEXP macro, which is defined as:
    #define MPFR_UEXP(X) (MPFR_ASSERTD ((X) >= 0), (mpfr_uexp_t) (X))

* If a mpfr_exp_t appears in arithmetic expressions together with ISO C90
  types int and/or long, computations must be done with the largest type,
  which is provided by mpfr_eexp_t.

* If a mpfr_exp_t needs to be converted from or to a MPFR number, the
  mpfr_set_exp_t or mpfr_get_exp_t macro should be used.

* If a mpfr_exp_t needs to be converted into a character string with a
  formatted output function (fprintf, printf, sprintf), the mpfr_eexp_t
  type should be used, together with the MPFR_EXP_FSPEC specifier, e.g.

    printf ("%" MPFR_EXP_FSPEC "d", (mpfr_eexp_t) exponent);

For implementation details, see the mpfr.h and mpfr-impl.h files.

===========================================================================

Be careful that the ternary value (a.k.a. "inexact") is not guaranteed
to be -1, 0, or 1, in general (for some functions, the exact value may
contain other information, such as midpoint cases with MPFR_EVEN_INEX),
and the exact behavior may change in the future. So, it is not correct
to multiply ternary values returned by arbitrary functions as this may
overflow.

To work with ternary values, mpfr-impl.h provides the following macros:

#define SIGN(I) ((I) < 0 ? -1 : (I) > 0)
#define SAME_SIGN(I1,I2) (SIGN (I1) == SIGN (I2))

===========================================================================

Because of a bug in the Mac OS X 10.5 linker, avoid tentative definitions
(C99, 6.9.2). Depending on the context, use either a simple declaration
(with the "extern" storage-class specifier) or an external definition.
This is also cleaner.

===========================================================================

In case of detected internal error, do not use printf() and exit().
Use assertions (MPFR_ASSERTN) instead.

===========================================================================

When using GNU extensions (based on the value of the __GNUC_* macros), check
whether they work with ICC. The following paper can give useful information:
"Intel® Compilers for Linux*: Compatibility with GNU Compilers" at
<https://software.intel.com/articles/intel-compilers-for-linux-compatibility-with-gnu-compilers>.

To detect compilers, see

  https://sourceforge.net/p/predef/wiki/Compilers/

===========================================================================

For configure tests, use AC_LINK_IFELSE rather than AC_COMPILE_IFELSE,
which is broken by design. The reason is that some errors just produce
a warning (which is not a bug from the compiler: in ISO C terminology,
this corresponds to a diagnostic, and the compilation is allowed to
succeed), and this is unfixable in a portable way.

===========================================================================

Shell portability
-----------------

Shell commands (in /bin/sh scripts, in Makefile and autotools related
files...) need to be valid in POSIX shells, but also in Bourne shells
(for instance, /bin/sh under Solaris is a Bourne shell).

In particular:

* Do not use $(...) but `...` (backticks).

* Be careful that quote nesting with backticks such as "`cmd "$foo"`"
  is not portable:

    https://unix.stackexchange.com/q/387246/74516
    ("quotes inside backticks inside quotes in ksh")

  But the external quotes are not needed when assigning to a variable:

    out=`cmd "$foo"`

  Otherwise one can write "`cmd \"$foo\"`".

===========================================================================

For developers - Use of the svn:eol-style Subversion property

The svn:eol-style Subversion property is traditionally set to "native" on
text files, but this has drawbacks:
  * On systems where the end-of-line (EOL) sequence is not LF, the obtained
    files are different from those from the tarballs. This makes maintenance
    harder.
  * Some tools under Windows (such as makeinfo of MinGW/MSYS) don't support
    the MS-Windows EOL sequence CRLF, and the MPFR build fails.

For these reasons, the svn:eol-style Subversion property should never be set
to "native".

===========================================================================

About the test suite
--------------------

When adding a test file for a new function (say mpfr_func), you can use
the following prototype tfunc.c (to put in the directory 'tests').
This file performs random tests for values of x between -5 and 5, with
a precision varying from 2 to 100.

You can add your own tests to this basic file. When adding the expected
result, do NOT use the one obtained from the MPFR function! Otherwise,
if this function is buggy, the test will be wrong and the function will
remain buggy. For random tests, avoid mpfr_urandomb as its values are
not truly random due to how it is specified (if the exponent is less
than 0, some of the trailing bits will necessarily be 0).

Do not forget to add 'tfunc' in the variable check_PROGRAMS
of the tests/Makefile.am file.

/* Test file for mpfr_func.

Copyright 2020 Free Software Foundation, Inc.
Contributed by the AriC and Caramba projects, INRIA.

This file is part of the GNU MPFR Library.

The GNU MPFR Library is free software; you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as published by
the Free Software Foundation; either version 3 of the License, or (at your
option) any later version.

The GNU MPFR Library is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
License for more details.

You should have received a copy of the GNU Lesser General Public License
along with the GNU MPFR Library; see the file COPYING.LESSER.  If not, see
https://www.gnu.org/licenses/ or write to the Free Software Foundation, Inc.,
51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. */

#include "mpfr-test.h"

#define TEST_FUNCTION mpfr_func
#define TEST_RANDOM_EMIN -5
#define TEST_RANDOM_EMAX 5
#include "tgeneric.c"

int
main (int argc, char *argv[])
{
  tests_start_mpfr ();

  test_generic (2, 100, 100);

  tests_end_mpfr ();
  return 0;
}

   ---------------------------------------------------------------------

Here is how the test suite works since the full Automake 1.13 support
(merge of the vl-am113 branch in r8821).

The tests_start_mpfr function, which should be called at the beginning
of each test program (unless nothing is tested and main() just contains
"return 77;"), starts by calling the test_version function, whose goal
is to do various header/library version checks of GMP and MPFR. In case
of mismatch between a header and a library, an error message is output
("make check" will redirect it to a log file). Then there are 3 cases:

1. An error in the MPFR version check is a fatal error: test_version()
   exits with an error (exit status = 1). The reason is that a different
   MPFR library (somewhere in some library search path) would probably
   be tested, so that the results of the test would be meaningless.

2. An error in the GMP version check is a non-fatal error: if there are
   no errors in MPFR version check, test_version() returns with value 1.
   However the tversion test program will regard this as a fatal error
   (thus "make check" will fail). The probable reason of the mismatch is
   that the GMP library has been upgraded while the MPFR test suite has
   not been rebuilt; otherwise there is probably something wrong in the
   GMP installation.

3. Otherwise test_version() returns with value 0 (everything is fine).

Note: The tests_start_mpfr function does a setbuf on stdout to disable
buffering. As a consequence, no operations on stdout (such as printf)
must be done before this function is called.

With Automake 1.13+, the tests are run in parallel if a -j make option
is used. In case of failure, information can be found in the log file
of each failed test program and in the global tests/test-suite.log file
(which is output automatically if the VERBOSE environment variable is
set to 1). If no tests fail, then the tests/tversion.log file is output
after the "testsuite summary"; it contains various useful information
about the MPFR build.

To use a wrapper to run the tests, such as valgrind or wine, define
LOG_COMPILER, e.g.:
  LOG_COMPILER="valgrind -q --error-exitcode=1 --leak-check=full" make check
  LOG_COMPILER=wine make check

More information about the parallel test harness:
https://www.gnu.org/software/automake/manual/automake.html#Parallel-Test-Harness

   ---------------------------------------------------------------------

In the tests, do not use `mpfr_set_d` (except when testing it), as the
result will depend on the floating-point arithmetic of the system;
this has shown many problems in the past and problems may still occur
with new systems. Use `mpfr_set_si` or `mpfr_set_str` instead.

To check the result of some function, use mpfr_equal_p rather than
mpfr_cmp, as mpfr_cmp will return 0 (equality) if the result is NaN.

Do not use functions that need optional features (except in a context
where such features are required). For instance, the mpfr_printf-like
functions need <stdarg.h> (HAVE_STDARG defined), thus should not be
used, except for testing them.

For temporary result files created by test programs, choose a unique
filename to avoid conflicts in parallel tests. To ensure that, the
filename should start with the name of the test program (for instance,
"tfprintf_out.txt" for tfprintf.c). Add the filename to CLEANFILES in
the tests/Makefile.am file.

Also, make sure that the tests run against previous MPFR versions,
possibly by disabling some tests with code like

  #if MPFR_VERSION >= MPFR_VERSION_NUM(2,3,0)

Indeed one can now easily run the trunk tests in a branch by executing

  svn switch .../svn/mpfr/trunk/tests tests

from the working copy. One can know when the tests directory has been
switched, thanks to

  $ svn status
      S  tests

In case of failure, freeing the memory explicitly is not necessary.
We do this in case of success just to be able to detect memory leaks
in MPFR.

   ---------------------------------------------------------------------

To check the coverage of the test suite, you can use gcov.
To get accurate information, do not enable optimizations.

   ./configure CFLAGS="--coverage"
   make clean
   make check
   find . -name '*.c' -exec gcov '{}' ';' | grep "lines executed" | sort

For each source file, there is a .c.gcov file which contains much more
information.

Another solution is to run the script 'coverage' within the 'tools' directory.

   ---------------------------------------------------------------------

To run the MPFR test suite under valgrind, you may need to do several
things.

First, as the running time is much longer than usual, you should not use
the --enable-tests-timeout configure option, or set the timeout value to
a large value; this can be done at run time, e.g. with

  export MPFR_TESTS_TIMEOUT=0

to disable the timeout, so that you do not need to rebuild MPFR for
this purpose.

Then just set the LOG_COMPILER environment variable to something like

  valgrind -q --error-exitcode=1 --leak-check=full

before running "make check", or type directly:

  LOG_COMPILER="valgrind -q --error-exitcode=1 --leak-check=full" make check

NOTE: with the new tests/Makefile.am file, the following is obsolete;
but it might still be useful under some occasions, e.g. if all the
valgrind output needs to be sent to a single file.

Before running valgrind, you should run "make check" a first time so
that everything is compiled out of valgrind.

Then you need the --trace-children=yes valgrind option (a possible
exception is when you run an individual test that has been built
statically). The reason is that libtool generates wrapper scripts
to link the tests against the right libraries. The drawback is that
you will get valgrind output for all the processes, including the
shell commands from the wrapper scripts (the --trace-children-skip
valgrind option will not allow you to filter every unwanted process).
But you can filter the output with:

  sed -n '/= Command: [^ ]*\/\.libs\/lt-/,/= ERROR SUMMARY:/p'

For readability, you should redirect the valgrind output to a file.
You can use --log-file, but due to --trace-children=yes, you need
the %p format specifier in the filename argument to generate a file
for each child; however many files will be generated, and it may be
better to use the following method to get a single file:

  valgrind --trace-children=yes --log-fd=3 make check 3> vg.out

then

  sed -n '/= Command: [^ ]*\/\.libs\/lt-/,/= ERROR SUMMARY:/p' vg.out

to get only the valgrind output corresponding to the MPFR tests.

Or if your shell supports it, you can use a process substitution
to filter the valgrind output directly to a file, e.g. with bash
or zsh:

  valgrind --trace-children=yes --log-fd=3 make check 3> >(sed -n \
    '/= Command: [^ ]*\/\.libs\/lt-/,/= ERROR SUMMARY:/p' > vg.out)

if you do not mind about the buffering delays.

   ---------------------------------------------------------------------

NOTE: with "AM_LDFLAGS = -no-install" in tests/Makefile.am, the following
is obsolete, as libtool no longer generates wrapper scripts; but it is left
here in case negative effects of "AM_LDFLAGS = -no-install" are seen or for
users with a special setup.

To debug some test program, e.g. tadd, with gdb, you cannot run "gdb tadd"
since libtool has generated a wrapper script to link the program against
the correct MPFR library. Instead, run:

  libtool --mode=execute gdb tadd

Alternatively, something like

  LD_PRELOAD=../src/.libs/libmpfr.so gdb .libs/tadd

may also work (example for GNU/Linux).

Note: for test programs not listed in Makefile.am (check_PROGRAMS),
libtool is not used (a static link against MPFR is done via LOADLIBES
in Makefile.am), so that gdb should be used in the conventional way.
You can use the following wrapper script to have a command that works
with both:

------------------------------------------------------------
#!/bin/sh

unset cmd
case $1 in
  -*) ;;
  ?*) test "x$(head -c 2 "$1")" = 'x#!' && \
      grep -q "^# Generated by libtool" "$1" && \
      cmd="libtool --mode=execute" ;;
esac

exec $cmd gdb "$@"
------------------------------------------------------------

and

  alias gdb='/path/to/the/wrapper/script'

===========================================================================

To cross-compile MPFR for i686-w64-mingw32 and test it under Wine:

0. Install wine (at least the 32-bit version).

1. Build and install GMP.

In the GMP source directory:
$ ./configure --host=i686-w64-mingw32 --disable-shared --prefix=...
$ make
$ make check LOG_COMPILER=wine
$ make install

Note: The -D__USE_MINGW_ANSI_STDIO option may be necessary in order to get
an ISO-compliant printf as mentioned in MPFR's INSTALL file. In this case,
add CC="i686-w64-mingw32-gcc -D__USE_MINGW_ANSI_STDIO" at the end of the
configure line. Otherwise MPFR needs to be configured with
  "CPPFLAGS=-DNPRINTF_J -DNPRINTF_L -DNPRINTF_T"
But with recent versions of the tools, printf now seems to be ISO-compliant
by default. The __USE_MINGW_ANSI_STDIO macro is now internal to MinGW and
should no longer be used.

2. Build and check MPFR.

In the MPFR source directory:
$ ./configure --host=i686-w64-mingw32 --disable-shared --with-gmp=...
$ make
$ make check LOG_COMPILER=wine

Note: Due to bugs in autoconf[1] and dash[2], the configure script
may create files with a binary filename or have any other arbitrary
behavior if /bin/sh is dash. The cause is that it tries to execute
a MS Windows executable, which is interpreted as a shell script by
dash (thus with random, meaningless commands). This will confuse
Subversion, and these files need to be removed manually.

[1] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=850329
[2] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=816313

===========================================================================

After a MPFR build, the list of GMP symbols used by this particular MPFR
build can be obtained as follows:

  nm -u src/.libs/libmpfr.so | sed -n 's/^ *U \(__gmp.*\)/\1/p'

at least under Linux, the library name and the "nm" behavior being
non-portable (adding the POSIX "-P" option may help, but there are
still differences between platforms).

Note that this list may depend on various parameters, such as the
architecture and the compilation options.

GMP internal symbols used by MPFR can be detected with the following
shell command (just replace /path/to/gmp.h by the actual pathname):

nm -u src/.libs/libmpfr.so | sed -n 's/^ *U \(__gmp.*\)/\1/p' | \
while read s
do
  case $s in
    __gmpn_*) regex="__MPN(${s#__gmpn_})" ;;
    *)        regex="$s" ;;
  esac
  grep -q "^#define .* ${regex}$" /path/to/gmp.h || echo "Internal: $s"
done

A similar check can be done with "make check-gmp-symbols".

One can also check that MPFR does not define exported symbols with a
prefix outside "mpfr_" and "__gmpfr_" by using "nm -g" and filtering
at least the "U" lines. But this can only be a manual check to avoid
false positives. Checking that a GMP reserved prefix is not used can
be done automatically, as with "make check-exported-symbols".

===========================================================================

To update the FAQ, checkout the misc directory of the repository root.
Modify the faq.xhtml file and run

  xsltproc --nodtdattr faq-web.xsl faq.xhtml > www/faq.html

Check with "svn diff" that this change has been done correctly (in case
of incorrect installation of XML tools), validate the files with

  xmllint --noout --loaddtd --valid faq.xhtml www/faq.html

and if everything is OK (no error messages), commit both files.

Update the FAQ.html file with update-faq in the doc directory of the
MPFR trunk and supported branches.

===========================================================================

Spelling:
  * Some suggestions: https://gcc.gnu.org/codingconventions.html#Spelling
  * Check with "codespell" (done by mpfrlint).