[PATCH kvm-unit-tests] configure: fix non-newlib cross-compiling

[PATCH kvm-unit-tests] configure: fix non-newlib cross-compiling

[Andrew Jones]

Commit 529046c3 "libcflat: add PRI(dux)32 format types" applies a
detection trick that requires native compiling or cross-compiling
with newlib, an embedded system C standard library. We can avoid
the new dependency with a different trick.

Cc: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Andrew Jones <drjones@redhat.com>
---
 configure | 14 +++++---------
 1 file changed, 5 insertions(+), 9 deletions(-)

diff --git a/configure b/configure
index 127868ce3090..3045dd129bda 100755
--- a/configure
+++ b/configure
@@ -12,6 +12,7 @@ host=$arch
 cross_prefix=
 endian=""
 pretty_print_stacks=yes
+u32_long=
 
 usage() {
     cat <<-EOF
@@ -110,16 +111,11 @@ if [ -f $testdir/run ]; then
 fi
 
 # check if uint32_t needs a long format modifier
-cat << EOF > lib_test.c
-#include <inttypes.h>
+cat << EOF > lib-test.c
+__UINT32_TYPE__
 EOF
-
-$cross_prefix$cc lib_test.c -E | grep "typedef" | grep "long" | grep "uint32_t" &> /dev/null
-exit=$?
-if [ $exit -eq 0 ]; then
-    u32_long=true
-fi
-rm -f lib_test.c
+u32_long=$($cross_prefix$cc -E lib-test.c | awk '! /^#/ && $2 == "long" {print "yes"}')
+rm -f lib-test.c
 
 # check for dependent 32 bit libraries
 if [ "$arch" != "arm" ]; then
-- 
2.9.3

Re: [PATCH kvm-unit-tests] configure: fix non-newlib cross-compiling

[Alex Bennée]

Andrew Jones <drjones@redhat.com> writes:

> Commit 529046c3 "libcflat: add PRI(dux)32 format types" applies a
> detection trick that requires native compiling or cross-compiling
> with newlib, an embedded system C standard library. We can avoid
> the new dependency with a different trick.
>
> Cc: Alex Bennée <alex.bennee@linaro.org>
> Signed-off-by: Andrew Jones <drjones@redhat.com>
> ---
>  configure | 14 +++++---------
>  1 file changed, 5 insertions(+), 9 deletions(-)
>
> diff --git a/configure b/configure
> index 127868ce3090..3045dd129bda 100755
> --- a/configure
> +++ b/configure
> @@ -12,6 +12,7 @@ host=$arch
>  cross_prefix=
>  endian=""
>  pretty_print_stacks=yes
> +u32_long=
>
>  usage() {
>      cat <<-EOF
> @@ -110,16 +111,11 @@ if [ -f $testdir/run ]; then
>  fi
>
>  # check if uint32_t needs a long format modifier
> -cat << EOF > lib_test.c
> -#include <inttypes.h>
> +cat << EOF > lib-test.c
> +__UINT32_TYPE__
>  EOF
> -
> -$cross_prefix$cc lib_test.c -E | grep "typedef" | grep "long" | grep "uint32_t" &> /dev/null
> -exit=$?
> -if [ $exit -eq 0 ]; then
> -    u32_long=true
> -fi
> -rm -f lib_test.c
> +u32_long=$($cross_prefix$cc -E lib-test.c | awk '! /^#/ && $2 == "long" {print "yes"}')
> +rm -f lib-test.c

Hmm this fails on my system as the -E output is:

  15:18 alex@overdrive01/aarch64  [kvm-unit-tests-32bit.git/review/drew-config@github] >arm-none-eabi-gcc -E test.c
  # 1 "test.c"
  # 1 "<built-in>"
  # 1 "<command-line>"
  # 1 "test.c"
  long unsigned int

Changing the line to:

  u32_long=$($cross_prefix$cc -E lib-test.c | awk '! /^#/ && $1 == "long" {print "yes"}')

Fixes the detection for me but I wonder if there are other corner cases
to catch?

The rest of the compile fails later due to un-merged format fixes for
pci_setup_msi which I guess is expected.

>
>  # check for dependent 32 bit libraries
>  if [ "$arch" != "arm" ]; then


--
Alex Bennée

Re: [PATCH kvm-unit-tests] configure: fix non-newlib cross-compiling

[Andrew Jones]

On Mon, Jan 16, 2017 at 03:22:00PM +0000, Alex Bennée wrote:
> 
> Andrew Jones <drjones@redhat.com> writes:
> 
> > Commit 529046c3 "libcflat: add PRI(dux)32 format types" applies a
> > detection trick that requires native compiling or cross-compiling
> > with newlib, an embedded system C standard library. We can avoid
> > the new dependency with a different trick.
> >
> > Cc: Alex Bennée <alex.bennee@linaro.org>
> > Signed-off-by: Andrew Jones <drjones@redhat.com>
> > ---
> >  configure | 14 +++++---------
> >  1 file changed, 5 insertions(+), 9 deletions(-)
> >
> > diff --git a/configure b/configure
> > index 127868ce3090..3045dd129bda 100755
> > --- a/configure
> > +++ b/configure
> > @@ -12,6 +12,7 @@ host=$arch
> >  cross_prefix=
> >  endian=""
> >  pretty_print_stacks=yes
> > +u32_long=
> >
> >  usage() {
> >      cat <<-EOF
> > @@ -110,16 +111,11 @@ if [ -f $testdir/run ]; then
> >  fi
> >
> >  # check if uint32_t needs a long format modifier
> > -cat << EOF > lib_test.c
> > -#include <inttypes.h>
> > +cat << EOF > lib-test.c
> > +__UINT32_TYPE__
> >  EOF
> > -
> > -$cross_prefix$cc lib_test.c -E | grep "typedef" | grep "long" | grep "uint32_t" &> /dev/null
> > -exit=$?
> > -if [ $exit -eq 0 ]; then
> > -    u32_long=true
> > -fi
> > -rm -f lib_test.c
> > +u32_long=$($cross_prefix$cc -E lib-test.c | awk '! /^#/ && $2 == "long" {print "yes"}')
> > +rm -f lib-test.c
> 
> Hmm this fails on my system as the -E output is:
> 
>   15:18 alex@overdrive01/aarch64  [kvm-unit-tests-32bit.git/review/drew-config@github] >arm-none-eabi-gcc -E test.c
>   # 1 "test.c"
>   # 1 "<built-in>"
>   # 1 "<command-line>"
>   # 1 "test.c"
>   long unsigned int
> 
> Changing the line to:
> 
>   u32_long=$($cross_prefix$cc -E lib-test.c | awk '! /^#/ && $1 == "long" {print "yes"}')

Let's change it to a couple simple greps then

 u32_long=$(gcc -E lib-test.c | grep -v '^#' | grep -q long && echo yes)

> 
> Fixes the detection for me but I wonder if there are other corner cases
> to catch?

Hope not, but I wouldn't be surprised. This type of detection is
pretty fragile...

> 
> The rest of the compile fails later due to un-merged format fixes for
> pci_setup_msi which I guess is expected.

Hmm, those are in master already. Did you apply this to a latest pull?

Thanks for testing! I'll send a v2.

drew

> 
> >
> >  # check for dependent 32 bit libraries
> >  if [ "$arch" != "arm" ]; then
> 
> 
> --
> Alex Bennée
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH kvm-unit-tests] configure: fix non-newlib cross-compiling

[Alex Bennée]

Andrew Jones <drjones@redhat.com> writes:

> On Mon, Jan 16, 2017 at 03:22:00PM +0000, Alex Bennée wrote:
>>
>> Andrew Jones <drjones@redhat.com> writes:
>>
>> > Commit 529046c3 "libcflat: add PRI(dux)32 format types" applies a
>> > detection trick that requires native compiling or cross-compiling
>> > with newlib, an embedded system C standard library. We can avoid
>> > the new dependency with a different trick.
>> >
>> > Cc: Alex Bennée <alex.bennee@linaro.org>
>> > Signed-off-by: Andrew Jones <drjones@redhat.com>
>> > ---
>> >  configure | 14 +++++---------
>> >  1 file changed, 5 insertions(+), 9 deletions(-)
>> >
>> > diff --git a/configure b/configure
>> > index 127868ce3090..3045dd129bda 100755
>> > --- a/configure
>> > +++ b/configure
>> > @@ -12,6 +12,7 @@ host=$arch
>> >  cross_prefix=
>> >  endian=""
>> >  pretty_print_stacks=yes
>> > +u32_long=
>> >
>> >  usage() {
>> >      cat <<-EOF
>> > @@ -110,16 +111,11 @@ if [ -f $testdir/run ]; then
>> >  fi
>> >
>> >  # check if uint32_t needs a long format modifier
>> > -cat << EOF > lib_test.c
>> > -#include <inttypes.h>
>> > +cat << EOF > lib-test.c
>> > +__UINT32_TYPE__
>> >  EOF
>> > -
>> > -$cross_prefix$cc lib_test.c -E | grep "typedef" | grep "long" | grep "uint32_t" &> /dev/null
>> > -exit=$?
>> > -if [ $exit -eq 0 ]; then
>> > -    u32_long=true
>> > -fi
>> > -rm -f lib_test.c
>> > +u32_long=$($cross_prefix$cc -E lib-test.c | awk '! /^#/ && $2 == "long" {print "yes"}')
>> > +rm -f lib-test.c
>>
>> Hmm this fails on my system as the -E output is:
>>
>>   15:18 alex@overdrive01/aarch64  [kvm-unit-tests-32bit.git/review/drew-config@github] >arm-none-eabi-gcc -E test.c
>>   # 1 "test.c"
>>   # 1 "<built-in>"
>>   # 1 "<command-line>"
>>   # 1 "test.c"
>>   long unsigned int
>>
>> Changing the line to:
>>
>>   u32_long=$($cross_prefix$cc -E lib-test.c | awk '! /^#/ && $1 == "long" {print "yes"}')
>
> Let's change it to a couple simple greps then
>
>  u32_long=$(gcc -E lib-test.c | grep -v '^#' | grep -q long && echo
>  yes)

Sounds good to me.

>
>>
>> Fixes the detection for me but I wonder if there are other corner cases
>> to catch?
>
> Hope not, but I wouldn't be surprised. This type of detection is
> pretty fragile...
>
>>
>> The rest of the compile fails later due to un-merged format fixes for
>> pci_setup_msi which I guess is expected.
>
> Hmm, those are in master already. Did you apply this to a latest pull?

Yeah I applied to tip. I think Paolo only took part of the merge but
that's OK I'll re-submit what's left in my series after I've gone
through comments.

>
> Thanks for testing! I'll send a v2.
>
> drew
>
>>
>> >
>> >  # check for dependent 32 bit libraries
>> >  if [ "$arch" != "arm" ]; then
>>
>>
>> --
>> Alex Bennée
>> --
>> To unsubscribe from this list: send the line "unsubscribe kvm" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html


--
Alex Bennée