build: cache the autoconf results of a package that opts in

A configure run spends its time on the compiler invocations behind its
checks, not on the shell. tools/m4 runs the compiler 633 times and takes
24 seconds. ccache does not help, because it refuses to cache an
autoconf conftest compile at all; it counts these separately as
"Autoconf compile/link" under its uncacheable calls.

A package that sets PKG_CONFIGURE_CACHE or HOST_CONFIGURE_CACHE keeps a
config.cache under tmp/, so a package that is cleaned and built again
reads the answers instead of running the checks. The configure stage of
tools/m4 drops from 24.5 s to 1.5 s. dirclean removes tmp/, so a full
reset of the tree starts from cold answers.

A package has to opt in, because a cache is only correct for a configure
script that keeps its side effects outside the AC_CACHE_VAL body.
ncurses appends -D_XOPEN_SOURCE and -D_DEFAULT_SOURCE to CPPFLAGS inside
that body, so a cached run restores the value, skips the append, and
then fails to compile wcwidth. ncurses_cfg.h is identical in both cases;
the difference is in the generated Makefile.

The path covers everything else that changes an answer. A target package
is keyed on the toolchain directory, which carries the gcc version. A
host package is keyed on the version string of the host compiler,
because autoconf cannot detect a new compiler behind an unchanged name;
gcc 14 turned an implicit function declaration into an error and changed
many answers that way. Both are keyed on a hash of the configure
arguments and the build flags, and autoconf refuses a cache written with
different flags on top of that.

A cached program path outlives a wipe of staging_dir, so ac_cv_path_ and
ac_cv_prog_ answers are dropped when the cache loads and configure
searches PATH again. Without that, a cache written before staging_dir
held sed pins /usr/bin/sed for good. A search runs no compiler, so this
costs nothing that matters.

The cache arrives through a site script rather than --cache-file, so a
hand written configure script never sees a new argument. The site script
removes the file of the previous arguments, so a different configuration
gets a cold cache instead of an autoconf error. It also unsets the
variable, and make exports it for the configure stage alone, so a nested
configure run keeps its own cache.

Set CONFIGURE_CACHE_DIR to keep the cache elsewhere, or to the empty
string to run every check again.

Host builds now read a site script on Linux as well. Before this change
only Darwin had one.

CONFIG_SITE holds one path name: autoconf before 2.70 reads the whole
value as that one name, so a second path would silence both scripts. It
names the cache script alone, and the cache script loads the site script
of the architecture from CONFIG_SITE_BASE.

Signed-off-by: John Crispin <john@phrozen.org>

include/host-build.mk

@@ -109,9 +109,12 @@ HOST_MAKE_FLAGS =
 
 HOST_CONFIGURE_CMD = $(BASH) ./configure
 
-ifeq ($(HOST_OS),Darwin)
-  HOST_CONFIG_SITE:=$(INCLUDE_DIR)/site/darwin
-endif
+HOST_CONFIG_SITE_BASE:=$(if $(filter Darwin,$(HOST_OS)),$(INCLUDE_DIR)/site/darwin)
+HOST_CONFIG_SITE:=$(INCLUDE_DIR)/site/cache
+# Set HOST_CONFIGURE_CACHE:=1 to keep the autoconf result cache of this host
+# build between builds. Read the note above PKG_CONFIGURE_CACHE in package.mk
+# before you opt a package in.
+HOST_CONFIGURE_CACHE_FILE = $(CONFIGURE_CACHE_BASE)/$(notdir $(BUILD_DIR_HOST))-$(host_cc_id)/$(notdir $(HOST_BUILD_DIR))/$(call strhash,$(HOST_CONFIGURE_ARGS) $(HOST_CONFIGURE_VARS)).cache
 
 define Host/Configure/Default
 	$(if $(HOST_CONFIGURE_PARALLEL),+)(cd $(HOST_BUILD_DIR)/$(3); \
@@ -170,6 +173,7 @@ define Host/Exports/Default
   $(1) : export PKG_CONFIG_LIBDIR=$$(HOST_BUILD_PREFIX)/lib/pkgconfig
   $(1) : export GIT_CEILING_DIRECTORIES=$$(BUILD_DIR_HOST)
   $(if $(HOST_CONFIG_SITE),$(1) : export CONFIG_SITE:=$(HOST_CONFIG_SITE))
+  $(if $(HOST_CONFIG_SITE_BASE),$(1) : export CONFIG_SITE_BASE:=$(HOST_CONFIG_SITE_BASE))
   $(if $(IS_PACKAGE_BUILD),$(1) : export PATH=$$(TARGET_PATH_PKG))
 endef
 Host/Exports=$(Host/Exports/Default)
@@ -192,6 +196,7 @@ ifndef DUMP
 	$(call BuildTimeLog,end,prepare)
 
   $(call Host/Exports,$(HOST_STAMP_CONFIGURED))
+  $(if $(and $(CONFIGURE_CACHE_BASE),$(HOST_CONFIGURE_CACHE)),$(HOST_STAMP_CONFIGURED) : export CONFIGURE_CACHE_FILE:=$(HOST_CONFIGURE_CACHE_FILE))
   $(HOST_STAMP_CONFIGURED): $(HOST_STAMP_PREPARED)
 	$(call BuildTimeLog,begin,configure)
 	$(foreach hook,$(Hooks/HostConfigure/Pre),$(call $(hook))$(sep))

include/package.mk

@@ -188,11 +188,19 @@ include $(INCLUDE_DIR)/autotools.mk
 _pkg_target:=$(if $(QUILT),,.)
 
 override MAKEFLAGS=
-CONFIG_SITE:=$(INCLUDE_DIR)/site/$(ARCH)
+CONFIG_SITE_BASE:=$(INCLUDE_DIR)/site/$(ARCH)
+CONFIG_SITE:=$(INCLUDE_DIR)/site/cache
+# Set PKG_CONFIGURE_CACHE:=1 to keep the autoconf result cache of this package
+# between builds. Opt in only after you check that the package configures the
+# same way with a warm cache as it does without one. A configure script that
+# changes CPPFLAGS or CFLAGS inside an AC_CACHE_VAL body is not safe: a warm
+# cache restores the value and skips the change. Compare the generated
+# Makefile, not only config.h, because that is where the difference shows.
+PKG_CONFIGURE_CACHE_FILE = $(CONFIGURE_CACHE_BASE)/$(TOOLCHAIN_DIR_NAME)/$(notdir $(PKG_BUILD_DIR))/$(call strhash,$(CONFIGURE_ARGS) $(CONFIGURE_VARS)).cache
 CUR_MAKEFILE:=$(filter-out Makefile,$(firstword $(MAKEFILE_LIST)))
 SUBMAKE:=$(NO_TRACE_MAKE) $(if $(CUR_MAKEFILE),-f $(CUR_MAKEFILE))
 PKG_CONFIG_PATH=$(STAGING_DIR)/usr/lib/pkgconfig:$(STAGING_DIR)/usr/share/pkgconfig
-unexport QUIET CONFIG_SITE
+unexport QUIET CONFIG_SITE CONFIG_SITE_BASE
 
 ifeq ($(DUMP)$(filter prereq clean refresh update,$(MAKECMDGOALS)),)
   ifneq ($(if $(QUILT),,$(CONFIG_AUTOREBUILD)),)
@@ -224,6 +232,7 @@ define Build/Exports/Default
   $(1) : export STAGING_PREFIX=$$(STAGING_DIR)/usr
   $(1) : export PATH=$$(TARGET_PATH_PKG)
   $(1) : export CONFIG_SITE:=$$(CONFIG_SITE)
+  $(1) : export CONFIG_SITE_BASE:=$$(CONFIG_SITE_BASE)
   $(1) : export PKG_CONFIG_PATH:=$$(PKG_CONFIG_PATH)
   $(1) : export PKG_CONFIG_LIBDIR:=$$(PKG_CONFIG_PATH)
   $(1) : export GIT_CEILING_DIRECTORIES:=$$(BUILD_DIR)
@@ -258,6 +267,7 @@ define Build/CoreTargets
 	$(call BuildTimeLog,end,prepare)
 
   $(call Build/Exports,$(STAMP_CONFIGURED))
+  $(if $(and $(CONFIGURE_CACHE_BASE),$(PKG_CONFIGURE_CACHE)),$(STAMP_CONFIGURED) : export CONFIGURE_CACHE_FILE:=$(PKG_CONFIGURE_CACHE_FILE))
   $(STAMP_CONFIGURED): $(STAMP_PREPARED) $(STAMP_CONFIGURED_DEPENDS)
 	$(call BuildTimeLog,begin,configure)
 	rm -f $(STAMP_CONFIGURED_WILDCARD)

include/site/cache

@@ -0,0 +1,44 @@
+#!/bin/sh
+#
+# Points the autoconf result cache at the file that the build system keeps for
+# this package. Set CONFIGURE_CACHE_DIR to switch this on; see rules.mk.
+#
+# configure sources the site scripts before it loads the cache, so setting
+# cache_file here has the same effect as --cache-file, without a configure
+# argument that a hand written configure script would choke on.
+
+# Autoconf before 2.70 reads CONFIG_SITE as a single path name, so the site
+# script of the architecture cannot be a second word in it. Load it from here.
+if [ -n "$CONFIG_SITE_BASE" ] && [ -r "$CONFIG_SITE_BASE" ]; then
+	. "$CONFIG_SITE_BASE"
+fi
+
+if [ -n "$CONFIGURE_CACHE_FILE" ]; then
+	configure_cache_dir=${CONFIGURE_CACHE_FILE%/*}
+
+	mkdir -p "$configure_cache_dir" 2>/dev/null
+
+	# A change of the configure arguments gives a new file name. Drop the
+	# file of the previous arguments so that the directory holds one cache
+	# per package.
+	for configure_cache_stale in "$configure_cache_dir"/*.cache; do
+		[ -f "$configure_cache_stale" ] || continue
+		[ "$configure_cache_stale" = "$CONFIGURE_CACHE_FILE" ] || rm -f "$configure_cache_stale"
+	done
+
+	# A cached program path outlives a wipe of staging_dir, so it can name a
+	# tool that is gone, or miss one that the build installed since. Drop
+	# those answers and let configure search PATH again. A search runs no
+	# compiler, and the compiler is where the time goes.
+	if [ -f "$CONFIGURE_CACHE_FILE" ]; then
+		grep -v -e '^ac_cv_path_' -e '^ac_cv_prog_' \
+			"$CONFIGURE_CACHE_FILE" > "$CONFIGURE_CACHE_FILE.tmp"
+		mv -f "$CONFIGURE_CACHE_FILE.tmp" "$CONFIGURE_CACHE_FILE"
+	fi
+
+	cache_file=$CONFIGURE_CACHE_FILE
+
+	# A configure script that runs further configure scripts must not hand
+	# this cache to them. Their checks run with different flags.
+	unset CONFIGURE_CACHE_FILE configure_cache_dir configure_cache_stale
+fi