The autoconf-based build system has room for improvements. The configure.ac script can be simplified in several places by using AS and AC macros or by defining new custom macros.

For example we have a lot of blocks that look like this:

AC_MSG_CHECKING(for chroot)
AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[#include <unistd.h>]], [[void x=chroot]])],
  [AC_DEFINE(HAVE_CHROOT, 1, Define if you have the 'chroot' function.)
   AC_MSG_RESULT(yes)],
  [AC_MSG_RESULT(no)
])

The block has an issue, too. It does not use AC_CACHE to cache the result. The entire block can be replaced by a custom macro that takes care of everything and implements correct caching:

  PY_CHECK_FUNC([chroot], [#include <unistd.h>])

We can also move several library and header checks from setup.py into configure.ac, e.g. check for soundcard.h or gdbm libs and headers.

The first PR adds helper macros, AC_CACHE_CHECK() [1] and AS_VAR_IF() [2]. It also unified internal variables to use format "ac_cv_func_$funcname", "ac_cv_func_lib_$library_$funcname", or "ac_cv_header_$headername_h". "ac_cv" stands for autoconf cached value.

AC_CACHE_CHECK() replaces AC_MSG_CHECKING() and AC_MSG_RESULT(). The syntax is AC_CACHE_CHECK([text], [cache variable], [body]) where body is only excecuted when the cache variable is not set. The body has to set the cache variable to yes or no. Any output and AC_DEFINE must occur outside the body.

AS_VAR_IF() is a nicer way to write if test $variable = value; then; fi.

[1] https://www.gnu.org/software/autoconf/manual/autoconf-2.69/html_node/Caching-Results.html#Caching-Results
[2] https://www.gnu.org/software/autoconf/manual/autoconf-2.69/html_node/Polymorphic-Variables.html

#73627 introduces forward compatibility issues with autoconf 2.71. I took the output of autoupdate and resolved all warnings.

New changeset be3cd5c by Christian Heimes in branch 'main':
bpo-45723: Detect missing pkg-config (GH-29442)
be3cd5c

New changeset 57c50c9 by Christian Heimes in branch 'main':
bpo-45723: Add helper macros and more caching to configure.ac (GH-29429)
57c50c9

New changeset 9bd0cf5 by Erlend Egeberg Aasland in branch 'main':
bpo-45723: Add macro for disabling/enabling CC warnings (GH-29466)
9bd0cf5

New changeset cbab997 by Christian Heimes in branch 'main':
bpo-45723: Prepare support for autoconf 2.71 (GH-29441)
cbab997

AC_CHECK_TYPE is obsolete and it's use is discouraged. There are some gotchas we need to check before switching to AC_CHECK_TYPES.

See:

• https://www.gnu.org/software/autoconf/manual/autoconf-2.70/html_node/Obsolete-Macros.html
• https://www.gnu.org/software/autoconf/manual/autoconf-2.70/html_node/Generic-Types.html

New changeset 1855336 by Erlend Egeberg Aasland in branch 'main':
bpo-45723: Remove obsolete AC_EXEEXT from configure.ac (GH-29486)
1855336

New changeset 49171aa by Erlend Egeberg Aasland in branch 'main':
bpo-45723: Remove dead code for obsolete --with-dyld option (GH-29500)
49171aa

New changeset 76d14fa by Erlend Egeberg Aasland in branch 'main':
bpo-45723: Improve and simplify more configure.ac checks (GH-29485)
76d14fa

New changeset fc9b622 by Christian Heimes in branch 'main':
bpo-45723: Add --with-pkg-config to configure (GH-29517)
fc9b622

New changeset db2277a by Erlend Egeberg Aasland in branch 'main':
bpo-45723: Add helpers for save/restore env (GH-29637)
db2277a

New changeset c1dec95 by Erlend Egeberg Aasland in branch 'main':
bpo-45723: Sort the grand AC_CHECK_HEADERS check (GH-29846)
c1dec95

New changeset 74b23c9 by Erlend Egeberg Aasland in branch 'main':
bpo-45723: Normalise configure user communication (GH-30024)
74b23c9

In the conversion to PY_CHECK_FUNC, there's a mistake in HAVE_EPOLL.

Python 3.10.1 defines HAVE_EPOLL by checking for the epoll_create function. Python 3.11.0a3 checks for the epoll function instead. There is no epoll() function so this always fails.

The effect is that epoll doesn't exist in the select module on Python 3.11.0a3. Most code that uses epoll falls back when it is not available, so this may not be failing any tests.

Thanks William!

New changeset 994f90c by Christian Heimes in branch 'main':
bpo-45723: Fix detection of epoll (bpo-30449)
994f90c

autoconf bumping to 2.71+ is a little annoying because LTS ubuntu 20.04 only ships with 2.69

should this have landed in 3.12 which is already in beta freeze? I'm building this for deadsnakes and it'd kind of suck to not ship 3.12 for ubuntu 20.04 especially when 3.12b1 built cleanly

autoconf bumping to 2.71+ is a little annoying because LTS ubuntu 20.04 only ships with 2.69

should this have landed in 3.12 which is already in beta freeze? I'm building this for deadsnakes and it'd kind of suck to not ship 3.12 for ubuntu 20.04 especially when 3.12b1 built cleanly

Perhaps you are misunderstanding something. The generated configure script is part of the sources; a clean build includes the generated configure script (try git clean -fdx).

IOW: You don't need GNU Autoconf at all for a clean build.

part of debian packaging is to regenerate configure

part of debian packaging is to regenerate configure

They can still do that with autoconf 2.69. There is no hard requirement to use 2.71. We only require 2.71 for CPython development. Debian should be all ok with 2.69.

the sources were updated to require 2.71 in #105207

the sources were updated to require 2.71 in #105207

For CPython development, yes.

Ah, there's a requirement check in the configure.ac script. Perhaps we can loosen that.

Ah, there's a requirement check in the configure.ac script. Perhaps we can loosen that.

yes that's the one I've been referring to

autoconf bumping to 2.71+ is a little annoying because LTS ubuntu 20.04 only ships with 2.69

should this have landed in 3.12 which is already in beta freeze? I'm building this for deadsnakes and it'd kind of suck to not ship 3.12 for ubuntu 20.04 especially when 3.12b1 built cleanly

Anthony: we've discussed this in the core dev group, and we reached the conclusion that we'll not be adjusting the hard requirement in the .ac file. For deadsnakes, I would suggest to simply patch the .ac file before regenerating configure. You should be fine with the following patch:

diff --git a/configure.ac b/configure.ac
index a24cd68973..793cd859ad 100644
--- a/configure.ac
+++ b/configure.ac
@@ -15,7 +15,7 @@ dnl
 # Set VERSION so we only need to edit in one place (i.e., here)
 m4_define([PYTHON_VERSION], [3.13])
 
-AC_PREREQ([2.71])
+AC_PREREQ([2.69])
 
 AC_INIT([python],[PYTHON_VERSION],[https://github.com/python/cpython/issues/])
 

I'll close this issue for now; Christian and I made several improvements to the AC code back in 3.10 and 3.11. IMO, future improvements should get their own targeted issues.

(cc. @tiran, in case you're around)

@erlend-aasland my fear is that 2.71+ syntax / features will get used -- what should I do then?

@erlend-aasland my fear is that 2.71+ syntax / features will get used -- what should I do then?

I suggest sticking to the recommendations of the devguide:

$ podman run --rm --pull=always -v $(pwd):/src:Z quay.io/tiran/cpython_autoconf:271
# ... or:
$ docker run --rm --pull=always -v $(pwd):/src quay.io/tiran/cpython_autoconf:271

Use the 269 tag iso. 271 for GNU Autoconf 2.69. Note also, that the autoconf version we use for 3.11 and older is not the stock GNU Autoconf v2.69; it is a patched version. So, once again, you're better off by using the devguide recommendations (or just use the configure script shipped with the source code) :)

I can't just do that unfortunately. that's outside my control as debian packages both must build from source and use the packages of the distribution

when 2.71+ syntax is used I'll basically have no choice but to drop the packages which is unfortunate but understandable

Well, on the bright side, no "2.71+ syntax" did sneak in yet, so deadsnakes should be fine for now :)