Remove package mpir

[mkoeppe]

MPIR, a fork of GMP, has not been updated upstream. The latest release, 3.0.0, is from 2017-03-01 (​https://mpir.org/downloads.html).

Sage currently provides an SPKG for MPIR as a configurable alternative to GMP. We should remove this option and package, as previously suggested in ​https://trac.sagemath.org/ticket/29620#comment:10

[tmonteil]

It seems there are some commits since then (the last is from December 2020 for Linux version and August 2021 for Windows version), perhaps should we ask to the people involved in MPIR how they see the future of MPIR ?

[mjo]

+1

Regardless of the state of MPIR upstream, sage never really developed a coherent plan for packages that are substitutes for one another. The ./configure options that are available now don't really make sense, and the best thing to do for our end users is to eliminate the confusion (if also the choice). If someone later wants to overhaul the build system to handle these choices consistently, then OK; but I don't think anyone does right now.

[dimpase]

The upstream (Bill) told me some time ago that due to Spectre and other CPU vulns discovered few years ago, parts of MPIR became obsolete, as speedups there relied on CPUs being unpatched.

IMHO it's semi-abandonware now.

[mkoeppe]

Replying to mjo:

sage never really developed a coherent plan for packages that are substitutes for one another. The ./configure options that are available now don't really make sense

#29620 tracks this

[mkoeppe]

New commits:

​9fa072e	build/pkgs/mpir: Remove
​81331d7	build/pkgs/yasm: Remove (was mpir dependency)

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​fabdbb3

build/pkgs/*/SPKG.rst: Clear out redundant 'Dependencies' sections

[jhpalmieri]

There are a few references to yasm: a comment in m4/sage_spkg_configure.m4 and many mentions in doc/en/developer/portability_testing.rst.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​724c3ff

src/doc/en/developer/portability_testing.rst: Remove references to package yasm

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​519dc83

m4/sage_spkg_configure.m4: In documentation, do not use yasm as an example

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​01406bf

build/pkgs/*/SPKG.rst: Clear out more redundant 'Dependencies' sections

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​205931c	src/doc/en/installation/source.rst: Remove doc of envvar SAGE_MP_LIBRARY
​e7fdb35	src/doc/en/installation/source.rst: Remove use of mpir as an example package

[mkoeppe]

In two places of the Sage library, also 'mpir' appears as a possible value of an algorithm parameter:

src/sage/arith/misc.py:        sage: a = binomial(100, 45, algorithm='mpir')
src/sage/rings/integer.pyx:    def binomial(self, m, algorithm='mpir'):

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​1eb83e3

build/pkgs/ratpoints/SPKG.rst: Remove redundant 'Dependencies' section

[dimpase]

Replying to mkoeppe:

In two places of the Sage library, also 'mpir' appears as a possible value of an algorithm parameter:

src/sage/arith/misc.py:        sage: a = binomial(100, 45, algorithm='mpir')
src/sage/rings/integer.pyx:    def binomial(self, m, algorithm='mpir'):

this option should be renamed gmp, and respective tests should drop # optional tag. Indeed, it just calls a gmp/mpir function (which are more or less the same, and mpir provided a replacement for gmp there).

sage: 100.binomial(45, algorithm='mpir')
61448471214136179596720592960

see, it still works, without any mpir installed.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​f883ea4

Integer.binomial: Add algorithm='gmp', keep 'mpir' as an alias

[mkoeppe]

Replying to dimpase:

this option should be renamed gmp

I have kept mpir as an alias, without deprecation warnings; after all, if we use system GMP, that could as well be MPIR

[dimpase]

there is still an #optional : mpir tag in integer.pyx, line 6427. And analogous gmp tag few lines down.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​1647e2f

src/sage/rings/integer.pyx: Rewrite a doctest so it works with gmp, mpir, 64bit, 32bit

[mkoeppe]

Replying to dimpase:

there is still an #optional : mpir tag in integer.pyx, line 6427. And analogous gmp tag few lines down.

These doctests weren't actually being run because neither mpir nor gmp were optional packages. I've rewritten it in a less specific way, it now runs

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​34526ca

build/pkgs/*/SPKG.rst: Remove some more redundant 'Dependencies' sections

[jhpalmieri]

There is a similar test in src/sage/ext/memory.pyx. We can also remove mpir from COPYING.txt. (Was there ever talk of autogenerating that file? Each package could have a file license.txt...)

[jhpalmieri]

Also in m4/sage_spkg_collect.m4:

    # Packages that should be included in the source distribution
    # This includes all standard packages and two special cases
    case "$SPKG_NAME" in
    mpir)
        in_sdist=yes
        ;;
    esac

[jhpalmieri]

I don't understand this comment, line 286 in src/sage/all.py:

    # Free globally allocated mpir integers.

Just delete the comment and leave the code? Replace the comment with something else? Remove the code?

[jhpalmieri]

Here are some proposed changes:

COPYING.txt

@@ -86 +86 @@
 mpc                         LGPLv3+
 mpfi                        COPYING is GPLv2, source files state LGPLv2.1+
 mpfr                        LGPLv3+
-mpir                        LGPLv3+
 mpmath                      Modified BSD
 networkx                    Modified BSD
 ntl                         GPLv2+

build/pkgs/ecm/spkg-configure.m4

@@ -1 +1 @@
 SAGE_SPKG_CONFIGURE([ecm], [
     m4_pushdef([SAGE_ECM_MINVER],[7.0.4])
     AC_REQUIRE([SAGE_SPKG_CONFIGURE_GMP])
-    AC_MSG_CHECKING([installing gmp/mpir? ])
+    AC_MSG_CHECKING([installing gmp? ])
     if test x$sage_spkg_install_gmp = xyes; then
         AC_MSG_RESULT([yes; install ecm as well])
         sage_spkg_install_ecm=yes

build/pkgs/gdb/SPKG.rst

@@ -19 +19 @@
 
 http://www.gnu.org/software/gdb/
 
-Dependencies
-------------
-
--  python
--  mpc
--  mpfr
--  ppl
--  gmp/mpir
--  makeinfo (external)
-
-
 Special Update/Build Instructions
 ---------------------------------
 

build/pkgs/isl/spkg-configure.m4

@@ -1 +1 @@
 SAGE_SPKG_CONFIGURE([isl], [
     AC_REQUIRE([SAGE_SPKG_CONFIGURE_GMP])
-    AC_MSG_CHECKING([installing gmp/mpir? ])
+    AC_MSG_CHECKING([installing gmp? ])
     if test x$sage_spkg_install_gmp = xyes; then
         AC_MSG_RESULT([yes; install isl as well])
         sage_spkg_install_isl=yes

build/pkgs/mpfr/spkg-configure.m4

@@ -1 +1 @@
 SAGE_SPKG_CONFIGURE([mpfr], [
     AC_REQUIRE([SAGE_SPKG_CONFIGURE_GMP])
-    AC_MSG_CHECKING([installing gmp/mpir? ])
+    AC_MSG_CHECKING([installing gmp? ])
     if test x$sage_spkg_install_gmp = xyes; then
         AC_MSG_RESULT([yes; install mpfr as well])
         sage_spkg_install_mpfr=yes

m4/sage_spkg_collect.m4

@@ -259 +259 @@
         AS_VAR_POPDEF([sage_require])dnl
         AS_VAR_POPDEF([sage_spkg_install])dnl
 
-    # Packages that should be included in the source distribution
-    # This includes all standard packages and two special cases
-    case "$SPKG_NAME" in
-    mpir)
-        in_sdist=yes
-        ;;
-    esac
-
     # Determine package source
     #
     if test -f "$DIR/requirements.txt"; then

[mkoeppe]

All looking fine, please push them to the ticket

[jhpalmieri]

Replying to mkoeppe:

All looking fine, please push them to the ticket

Done

---

New commits:

​e3ff6aa

trac 32549: misc small changes, plus don't include mpir in sdist

[dimpase]

Let's use this ticket to update to consistently use SAGE_SPKG_DEPCHECK instead of that if test ... in all the packages affected.

[dimpase]

the monster isn't completely gone yet:

build//pkgs/ppl/spkg-install.in:# Make sure that we prefer Sage's mpir library over system-wide gmp/mpir installs
src/doc/it/faq/faq-usage.rst:    rm spkg/installed/mpir* spkg/installed/atlas*
src/doc/it/faq/faq-general.rst:`MPIR <http://www.mpir.org>`_,
src/doc/en/faq/faq-usage.rst:    $ rm spkg/installed/mpir* spkg/installed/atlas*
src/doc/en/faq/faq-general.rst:`MPIR <http://www.mpir.org>`_,
src/sage/ext_data/valgrind/sage-additional.supp:   fun:sage_mpir_malloc
src/sage/ext_data/valgrind/sage-additional.supp:   mpir_realloc
src/sage/ext_data/valgrind/sage-additional.supp:   fun:sage_mpir_realloc
src/sage/ext/memory.pyx:    sage: 2^(2^63-2)      # optional - mpir
src/sage/all.py:    # Free globally allocated mpir integers.

taking care of this now.

---

New commits:

​ccd0ff5

use SAGE_SPKG_DEPCHECK

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​e400f0e	makes gmp test unconditional, remove mpir test
​07da6c3	fix mpir mentions in FAQs and source comments

[mkoeppe]

Looking good. Let's merge #30350 (ATLAS removal) to resolve merge conflicts in the documentation and other places

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​9ed04f7

Merge tag '9.5.beta2' into t/32549/remove_package_mpir

[dimpase]

there were a couple of docbuild errors in FAQs in my branch. Should merge the latest commit from u/dimpase/remove_package_mpir

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​8c071a8	https-fy consistently
​7c84f31	Merge branch 'u/dimpase/remove_package_mpir' of git://trac.sagemath.org/sage into t/32549/remove_package_mpir

[dimpase]

OK, this works.

[vbraun]

Merge conflict

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​2576f86	build/make/Makefile.in: If a script package has no spkg-install, run "sage -info" and exit with error
​feb8de7	build/pkgs/: Remove spkg-install scripts for dummy script packages
​8f782c0	.github/workflows/tox-{optional,experimental}.yml: Do not try to test dummy script packages
​4b292be	build/pkgs/perl_mongodb/spkg-install: Remove
​a7b6352	build/bin/sage-spkg-info: Fix display of system packages
​e65b309	bootstrap: Do not provide ./configure --enable-SPKG options for dummy optional packages
​b485d46	m4/sage_spkg_collect.m4: Do not advertise dummy optional packages as installable
​286b39c	Merge #31163

[was]

perhaps should we ask to the people involved in MPIR how they see the future of MPIR ?

Bill also told me recently that MPIR is not going forward at this point for a combination of reasons. +1 to this ticket.