[gentoo-dev] [PATCHES] multilib-build: use MULTILIB_ABI for eclass-specific ABI value

[gentoo-dev] [PATCHES] multilib-build: use MULTILIB_ABI for eclass-specific ABI value

[Michał Górny]

[-- Attachment #1: Type: text/plain, Size: 1104 bytes --]

Hello,

Currently, the multilib-build eclass uses abi_* constants only for USE
flags and only ${ABI} is exported to the function. This is bad since it
basically requires a reverse mapping of ABI->abi_* values, often
inlined as ${ABI} checks.

The patches which I will send in reply to this thread aim to fix it.

The first patch changes the eclass logic. The abi_* values, with 'abi_'
prefix stripped, are called MULTILIB_ABI now. They are used to run
the 'foreach' functions, and now are set in the called functions along
with ABI.

As a downside, the switch required the MULTILIB_ABI -> ABI mapping to
occur inside foreach -- as in, another 'for' loop. It shouldn't cause
any noticeable difference.

Additionally, the 'default' fallback no longer calls
multilib_toolchain_setup. This should improve compatibility with
multilib-portage and *maybe* cross-compiling.

The second patch uses new ${MULTILIB_ABI} variable in header wrapping
function. In other words, it removes the ABI -> MULTILIB_ABI reverse
mapping which becomes unnecessary.

-- 
Best regards,
Michał Górny

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 966 bytes --]

[gentoo-dev] [PATCH 1/2] multilib-build: set MULTILIB_ABI as the 'global' ABI value.

[Michał Górny]

---
 gx86/eclass/multilib-build.eclass | 41 ++++++++++++++++++++++++++++++++++-----
 1 file changed, 36 insertions(+), 5 deletions(-)

diff --git a/gx86/eclass/multilib-build.eclass b/gx86/eclass/multilib-build.eclass
index dbaed70..fdaed6b 100644
--- a/gx86/eclass/multilib-build.eclass
+++ b/gx86/eclass/multilib-build.eclass
@@ -25,10 +25,22 @@ esac
 
 inherit multibuild multilib
 
+# @ECLASS-VARIABLE: MULTILIB_ABI
+# @DEFAULT-UNSET
+# @DESCRIPTION:
+# The current ABI, in form of base arch and sub-ABI, joined using
+# an underscore. It's equal to the flag name with 'abi_' prefix
+# stripped.
+#
+# Exported by multilib_foreach_abi, multilib_parallel_foreach_abi,
+# multilib_for_best_abi.
+#
+# Example values: x86_32, mips_n32
+
 # @ECLASS-VARIABLE: _MULTILIB_FLAGS
 # @INTERNAL
 # @DESCRIPTION:
-# The list of multilib flags and corresponding ABI values.
+# The list of corresponding USE flags and ABI values.
 _MULTILIB_FLAGS=(
 	abi_x86_32:x86
 	abi_x86_64:amd64
@@ -56,6 +68,7 @@ _multilib_build_set_globals() {
 _multilib_build_set_globals
 
 # @FUNCTION: multilib_get_enabled_abis
+# @INTERNAL
 # @DESCRIPTION:
 # Return the ordered list of enabled ABIs if multilib builds
 # are enabled. The best (most preferred) ABI will come last.
@@ -74,7 +87,7 @@ multilib_get_enabled_abis() {
 			local m_flag=${i%:*}
 
 			if [[ ${m_abi} == ${abi} ]] && use "${m_flag}"; then
-				echo "${abi}"
+				echo "${m_flag#abi_}"
 				found=1
 			fi
 		done
@@ -87,7 +100,7 @@ multilib_get_enabled_abis() {
 
 		debug-print "${FUNCNAME}: no ABIs enabled, fallback to ${abi}"
 		debug-print "${FUNCNAME}: ABI=${ABI}, DEFAULT_ABI=${DEFAULT_ABI}"
-		echo ${abi}
+		echo default
 	fi
 }
 
@@ -99,8 +112,26 @@ multilib_get_enabled_abis() {
 _multilib_multibuild_wrapper() {
 	debug-print-function ${FUNCNAME} "${@}"
 
-	local ABI=${MULTIBUILD_VARIANT}
-	multilib_toolchain_setup "${ABI}"
+	local MULTILIB_ABI=${MULTIBUILD_VARIANT}
+	local i
+
+	if [[ ${MULTILIB_ABI} != default ]]; then
+		local ABI
+
+		for i in "${_MULTILIB_FLAGS[@]}"; do
+			local m_abi=${i#*:}
+			local m_flag=${i%:*}
+
+			if [[ ${MULTILIB_ABI} == ${m_flag#abi_} ]]; then
+				local -x ABI=${m_abi}
+				break
+			fi
+		done
+
+		[[ ${ABI} ]] || die "Unable to match ${MULTILIB_ABI} to an ABI!"
+		multilib_toolchain_setup "${ABI}"
+	fi
+
 	"${@}"
 }
 
-- 
1.8.1.5

[gentoo-dev] [PATCH 2/2] Use MULTILIB_ABI in header wrapping code.

[Michał Górny]

---
 gx86/eclass/autotools-multilib.eclass | 21 ++++-----------------
 1 file changed, 4 insertions(+), 17 deletions(-)

diff --git a/gx86/eclass/autotools-multilib.eclass b/gx86/eclass/autotools-multilib.eclass
index 5ecbd2f..55d32d7 100644
--- a/gx86/eclass/autotools-multilib.eclass
+++ b/gx86/eclass/autotools-multilib.eclass
@@ -102,33 +102,20 @@ _autotools-multilib_wrap_headers() {
 
 #if defined(__x86_64__) /* amd64 */
 #	if defined(__ILP32__) /* x32 ABI */
-#		error "abi_x86_x32 not supported by the package."
+#		error "x86_x32 not supported by the package."
 #	else /* 64-bit ABI */
-#		error "abi_x86_64 not supported by the package."
+#		error "x86_64 not supported by the package."
 #	endif
 #elif defined(__i386__) /* plain x86 */
-#	error "abi_x86_32 not supported by the package."
+#	error "x86_32 not supported by the package."
 #else
 #	error "No ABI matched, please report a bug to bugs.gentoo.org"
 #endif
 _EOF_
 		fi
 
-		# XXX: get abi_* directly
-		local abi_flag
-		case "${ABI}" in
-			amd64)
-				abi_flag=abi_x86_64;;
-			x86)
-				abi_flag=abi_x86_32;;
-			x32)
-				abi_flag=abi_x86_x32;;
-			*)
-				die "Header wrapping for ${ABI} not supported yet";;
-		esac
-
 		# Note: match a space afterwards to avoid collision potential.
-		sed -e "/${abi_flag} /s&error.*&include <${CHOST}/${f}>&" \
+		sed -e "/${MULTILIB_ABI} /s&error.*&include <${CHOST}/${f}>&" \
 			-i "${ED}/tmp/multilib-include${f}" || die
 	done
 }
-- 
1.8.1.5

Re: [gentoo-dev] [PATCHES] multilib-build: use MULTILIB_ABI for eclass-specific ABI value

[Thomas Sachau]

[-- Attachment #1: Type: text/plain, Size: 1160 bytes --]

Michał Górny schrieb:
> Hello,
> 
> Currently, the multilib-build eclass uses abi_* constants only for USE
> flags and only ${ABI} is exported to the function. This is bad since it
> basically requires a reverse mapping of ABI->abi_* values, often
> inlined as ${ABI} checks.
> 
> The patches which I will send in reply to this thread aim to fix it.
> 
> The first patch changes the eclass logic. The abi_* values, with 'abi_'
> prefix stripped, are called MULTILIB_ABI now. They are used to run
> the 'foreach' functions, and now are set in the called functions along
> with ABI.
> 
> As a downside, the switch required the MULTILIB_ABI -> ABI mapping to
> occur inside foreach -- as in, another 'for' loop. It shouldn't cause
> any noticeable difference.
> 
> Additionally, the 'default' fallback no longer calls
> multilib_toolchain_setup. This should improve compatibility with
> multilib-portage and *maybe* cross-compiling.

You know, that multilib-portage does use MULTILIB_ABI as USE-expanded
variable? Using exactly the same in the eclass will call for collision
issues.


-- 

Thomas Sachau
Gentoo Linux Developer


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 379 bytes --]

Re: [gentoo-dev] [PATCHES] multilib-build: use MULTILIB_ABI for eclass-specific ABI value

[Ciaran McCreesh]

[-- Attachment #1: Type: text/plain, Size: 364 bytes --]

On Wed, 03 Apr 2013 11:40:31 +0200
Thomas Sachau <tommy@gentoo.org> wrote:
> You know, that multilib-portage does use MULTILIB_ABI as USE-expanded
> variable? Using exactly the same in the eclass will call for collision
> issues.

I doubt very many people know that, since there's still no spec for the
changes in multilib-portage.

-- 
Ciaran McCreesh

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: [gentoo-dev] [PATCHES] multilib-build: use MULTILIB_ABI for eclass-specific ABI value

[Michał Górny]

[-- Attachment #1: Type: text/plain, Size: 1321 bytes --]

On Wed, 03 Apr 2013 11:40:31 +0200
Thomas Sachau <tommy@gentoo.org> wrote:

> Michał Górny schrieb:
> > Hello,
> > 
> > Currently, the multilib-build eclass uses abi_* constants only for USE
> > flags and only ${ABI} is exported to the function. This is bad since it
> > basically requires a reverse mapping of ABI->abi_* values, often
> > inlined as ${ABI} checks.
> > 
> > The patches which I will send in reply to this thread aim to fix it.
> > 
> > The first patch changes the eclass logic. The abi_* values, with 'abi_'
> > prefix stripped, are called MULTILIB_ABI now. They are used to run
> > the 'foreach' functions, and now are set in the called functions along
> > with ABI.
> > 
> > As a downside, the switch required the MULTILIB_ABI -> ABI mapping to
> > occur inside foreach -- as in, another 'for' loop. It shouldn't cause
> > any noticeable difference.
> > 
> > Additionally, the 'default' fallback no longer calls
> > multilib_toolchain_setup. This should improve compatibility with
> > multilib-portage and *maybe* cross-compiling.
> 
> You know, that multilib-portage does use MULTILIB_ABI as USE-expanded
> variable? Using exactly the same in the eclass will call for collision
> issues.

Argv. Could you suggest a new name then?

-- 
Best regards,
Michał Górny

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 966 bytes --]

Re: [gentoo-dev] [PATCHES] multilib-build: use MULTILIB_ABI for eclass-specific ABI value

[Michał Górny]

[-- Attachment #1: Type: text/plain, Size: 1016 bytes --]

On Wed, 03 Apr 2013 11:40:31 +0200
Thomas Sachau <tommy@gentoo.org> wrote:

> Michał Górny schrieb:
> > The first patch changes the eclass logic. The abi_* values, with 'abi_'
> > prefix stripped, are called MULTILIB_ABI now. They are used to run
> > the 'foreach' functions, and now are set in the called functions along
> > with ABI.
> >
> > [...]
> > 
> > Additionally, the 'default' fallback no longer calls
> > multilib_toolchain_setup. This should improve compatibility with
> > multilib-portage and *maybe* cross-compiling.
> 
> You know, that multilib-portage does use MULTILIB_ABI as USE-expanded
> variable? Using exactly the same in the eclass will call for collision
> issues.

Ok, I see one more issue. Whenever ebuilds start checking
MULTILIB_ABI or the new-named equivalent of it, multilib portage is
going to fail those checks by using 'default'.

Any suggestions how to avoid that? Probably achievable only through
staying with $ABI.

-- 
Best regards,
Michał Górny

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 966 bytes --]

Re: [gentoo-dev] [PATCHES] multilib-build: use MULTILIB_ABI for eclass-specific ABI value

[Michał Górny]

[-- Attachment #1: Type: text/plain, Size: 728 bytes --]

On Tue, 2 Apr 2013 23:42:42 +0200
Michał Górny <mgorny@gentoo.org> wrote:

> Currently, the multilib-build eclass uses abi_* constants only for USE
> flags and only ${ABI} is exported to the function. This is bad since it
> basically requires a reverse mapping of ABI->abi_* values, often
> inlined as ${ABI} checks.

Due to the multilib-portage conflicts pointed out by Tommy, and other
potential issues resulting from spreading the use of abi_* constants
over the tree, I have decided to withdraw these patches for now.

They may be re-considered in the future when multilib-portage will
respect and use the eclass, or when a better and less conflicting
method is found.

-- 
Best regards,
Michał Górny

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 966 bytes --]