[gentoo-dev] [PATCH] toolchain-funcs.eclass: avoid using tc-getCPP in compiler checks

[gentoo-dev] [PATCH] toolchain-funcs.eclass: avoid using tc-getCPP in compiler checks

[Mike Gilbert]

We usually want to invoke cc -E rather than cpp directly. We should not
rely on the quirk of tc-getCPP defaulting to "cc -E" when CPP is unset.

Resolves a warning when CPP is set to llvm-cpp, which does not accept
the -E option.

Closes: https://bugs.gentoo.org/949340
Signed-off-by: Mike Gilbert <floppym@gentoo.org>
---
 eclass/toolchain-funcs.eclass | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/eclass/toolchain-funcs.eclass b/eclass/toolchain-funcs.eclass
index 0abed5b8d75e..754d23f91c06 100644
--- a/eclass/toolchain-funcs.eclass
+++ b/eclass/toolchain-funcs.eclass
@@ -1,4 +1,4 @@
-# Copyright 2002-2024 Gentoo Authors
+# Copyright 2002-2025 Gentoo Authors
 # Distributed under the terms of the GNU General Public License v2
 
 # @ECLASS: toolchain-funcs.eclass
@@ -900,7 +900,7 @@ tc-get-compiler-type() {
 	HAVE_GCC
 #endif
 '
-	local res=$($(tc-getCPP "$@") -E -P - <<<"${code}")
+	local res=$($(tc-getCC "$@") -E -P - <<<"${code}")
 
 	case ${res} in
 		*HAVE_PATHCC*)	echo pathcc;;
@@ -927,7 +927,7 @@ tc-is-clang() {
 # compilers rather than maintaining a --version flag matrix, bug #335943.
 _gcc_fullversion() {
 	local ver="$1"; shift
-	set -- $($(tc-getCPP "$@") -E -P - <<<"__GNUC__ __GNUC_MINOR__ __GNUC_PATCHLEVEL__")
+	set -- $($(tc-getCC "$@") -E -P - <<<"__GNUC__ __GNUC_MINOR__ __GNUC_PATCHLEVEL__")
 	eval echo "${ver}"
 }
 
@@ -960,7 +960,7 @@ gcc-micro-version() {
 # Internal func. Based on _gcc_fullversion() above.
 _clang_fullversion() {
 	local ver="$1"; shift
-	set -- $($(tc-getCPP "$@") -E -P - <<<"__clang_major__ __clang_minor__ __clang_patchlevel__")
+	set -- $($(tc-getCC "$@") -E -P - <<<"__clang_major__ __clang_minor__ __clang_patchlevel__")
 	eval echo "${ver}"
 }
 
@@ -1347,14 +1347,14 @@ tc-get-c-rtlib() {
 # @FUNCTION: tc-get-ptr-size
 # @RETURN: Size of a pointer in bytes for CHOST (e.g. 4 or 8).
 tc-get-ptr-size() {
-	$(tc-getCPP) -P - <<< __SIZEOF_POINTER__ ||
+	$(tc-getCC) -E -P - <<< __SIZEOF_POINTER__ ||
 		die "Could not determine CHOST pointer size"
 }
 
 # @FUNCTION: tc-get-build-ptr-size
 # @RETURN: Size of a pointer in bytes for CBUILD (e.g. 4 or 8).
 tc-get-build-ptr-size() {
-	$(tc-getBUILD_CPP) -P - <<< __SIZEOF_POINTER__ ||
+	$(tc-getBUILD_CC) -E -P - <<< __SIZEOF_POINTER__ ||
 		die "Could not determine CBUILD pointer size"
 }
 
-- 
2.48.1

Re: [gentoo-dev] [PATCH] toolchain-funcs.eclass: avoid using tc-getCPP in compiler checks

[James Le Cuirot]

[-- Attachment #1: Type: text/plain, Size: 2674 bytes --]

On Wed, 2025-02-05 at 13:10 -0500, Mike Gilbert wrote:
> We usually want to invoke cc -E rather than cpp directly. We should not
> rely on the quirk of tc-getCPP defaulting to "cc -E" when CPP is unset.
> 
> Resolves a warning when CPP is set to llvm-cpp, which does not accept
> the -E option.
> 
> Closes: https://bugs.gentoo.org/949340
> Signed-off-by: Mike Gilbert <floppym@gentoo.org>
> ---
>  eclass/toolchain-funcs.eclass | 12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)
> 
> diff --git a/eclass/toolchain-funcs.eclass b/eclass/toolchain-funcs.eclass
> index 0abed5b8d75e..754d23f91c06 100644
> --- a/eclass/toolchain-funcs.eclass
> +++ b/eclass/toolchain-funcs.eclass
> @@ -1,4 +1,4 @@
> -# Copyright 2002-2024 Gentoo Authors
> +# Copyright 2002-2025 Gentoo Authors
>  # Distributed under the terms of the GNU General Public License v2
>  
>  # @ECLASS: toolchain-funcs.eclass
> @@ -900,7 +900,7 @@ tc-get-compiler-type() {
>  	HAVE_GCC
>  #endif
>  '
> -	local res=$($(tc-getCPP "$@") -E -P - <<<"${code}")
> +	local res=$($(tc-getCC "$@") -E -P - <<<"${code}")
>  
>  	case ${res} in
>  		*HAVE_PATHCC*)	echo pathcc;;
> @@ -927,7 +927,7 @@ tc-is-clang() {
>  # compilers rather than maintaining a --version flag matrix, bug #335943.
>  _gcc_fullversion() {
>  	local ver="$1"; shift
> -	set -- $($(tc-getCPP "$@") -E -P - <<<"__GNUC__ __GNUC_MINOR__ __GNUC_PATCHLEVEL__")
> +	set -- $($(tc-getCC "$@") -E -P - <<<"__GNUC__ __GNUC_MINOR__ __GNUC_PATCHLEVEL__")
>  	eval echo "${ver}"
>  }
>  
> @@ -960,7 +960,7 @@ gcc-micro-version() {
>  # Internal func. Based on _gcc_fullversion() above.
>  _clang_fullversion() {
>  	local ver="$1"; shift
> -	set -- $($(tc-getCPP "$@") -E -P - <<<"__clang_major__ __clang_minor__ __clang_patchlevel__")
> +	set -- $($(tc-getCC "$@") -E -P - <<<"__clang_major__ __clang_minor__ __clang_patchlevel__")
>  	eval echo "${ver}"
>  }
>  
> @@ -1347,14 +1347,14 @@ tc-get-c-rtlib() {
>  # @FUNCTION: tc-get-ptr-size
>  # @RETURN: Size of a pointer in bytes for CHOST (e.g. 4 or 8).
>  tc-get-ptr-size() {
> -	$(tc-getCPP) -P - <<< __SIZEOF_POINTER__ ||
> +	$(tc-getCC) -E -P - <<< __SIZEOF_POINTER__ ||
>  		die "Could not determine CHOST pointer size"
>  }
>  
>  # @FUNCTION: tc-get-build-ptr-size
>  # @RETURN: Size of a pointer in bytes for CBUILD (e.g. 4 or 8).
>  tc-get-build-ptr-size() {
> -	$(tc-getBUILD_CPP) -P - <<< __SIZEOF_POINTER__ ||
> +	$(tc-getBUILD_CC) -E -P - <<< __SIZEOF_POINTER__ ||
>  		die "Could not determine CBUILD pointer size"
>  }
>  

I'm slightly unsure, but this has always confused me. If it works for you...

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 858 bytes --]

Re: [gentoo-dev] [PATCH] toolchain-funcs.eclass: avoid using tc-getCPP in compiler checks

[Ionen Wolkens]

[-- Attachment #1: Type: text/plain, Size: 1187 bytes --]

On Wed, Feb 05, 2025 at 01:10:10PM -0500, Mike Gilbert wrote:
> We usually want to invoke cc -E rather than cpp directly. We should not
> rely on the quirk of tc-getCPP defaulting to "cc -E" when CPP is unset.

I think it's a good change, but just to note it could break things
that relied on CPP being what's used.

Aka these don't have a proper way to say "test this compiler" and,
given it uses CPP, it's what I ended up setting:

    CPP="${KERNEL_CC} -E" tc-is-clang ...

That's from nvidia-drivers ebuilds, which I'll fix real soon so no
need to worry about these. But I don't know what else may have done
this, not that I think it's going to be common.

I did something similar in linux-mod-r1.eclass but I took more
precaution there so it's not affected:

    # for toolchain-funcs, uses CPP > CC but set both not to make assumptions
    local CC=${KERNEL_CC} CPP="${KERNEL_CC} -E" LD=${KERNEL_LD}

I could cleanup the CPP bit if this is merged though. That it doesn't
use CC was confusing if anything thus needing a comment to explain
why it's being set.

Hopefully haven't forgot anything else I did, think wine & mingw things
are fine.
-- 
ionen

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [gentoo-dev] [PATCH] toolchain-funcs.eclass: avoid using tc-getCPP in compiler checks

[Mike Gilbert]

On Wed, Feb 5, 2025 at 2:06 PM Ionen Wolkens <ionen@gentoo.org> wrote:
>
> On Wed, Feb 05, 2025 at 01:10:10PM -0500, Mike Gilbert wrote:
> > We usually want to invoke cc -E rather than cpp directly. We should not
> > rely on the quirk of tc-getCPP defaulting to "cc -E" when CPP is unset.
>
> I think it's a good change, but just to note it could break things
> that relied on CPP being what's used.

Thanks for that reminder.