Skip to contentSkip to conversation

python: fix double format detection

PATCH New Jérôme Pouiller 3 participants 4 messages Nov 9, 2016

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 python: fix double format detection

Python is not able to detect if compiler double representation is compliant with IEE754:

checking whether C doubles are little-endian IEEE 754 binary64... no checking whether C doubles are big-endian IEEE 754 binary64... no checking whether C doubles are ARM mixed-endian IEEE 754 binary64... no

Accordingly 'legacy' mode isused. It is possible to check this at runtime by check if 'sys.float_repr_style' contains 'short' or 'legacy'. Calculus correctness is not garanteed with 'legacy'.

Problem is better described here:

http://stackoverflow.com/questions/29920294/what-causes-pythons-float-repr-style-to-use-legacy
https://bugs.python.org/issue7117

However, all gcc architecture use a representation compliant with IEE754. So, we can enable it unconditionnaly.

---
 package/python/python.mk | 7 +++++++
 1 file changed, 7 insertions(+)
package/python/python.mk +7 -0
index cc65376..0191d59 100644
@@ -140,6 +140,13 @@ PYTHON_CONF_ENV += \
 	ac_cv_prog_HAS_HG=/bin/false \
 	ac_cv_prog_SVNVERSION=/bin/false
 
+# GCC is always complient with IEEE754
+ifeq ($(call qstrip,$(BR2_ENDIAN)),LITTLE)
+PYTHON_CONF_ENV += ac_cv_little_endian_double=yes
+else
+PYTHON_CONF_ENV += ac_cv_big_endian_double=yes
+endif
+
 PYTHON_CONF_OPTS += \
 	--without-cxx-main 	\
 	--without-doc-strings	\
Signature · 1 line
2.9.3
#1 lore raw

PATCH 2/2 python3: fix double format detection

Python is not able to detect if compiler double representation is compliant with IEE754:

checking whether C doubles are little-endian IEEE 754 binary64... no checking whether C doubles are big-endian IEEE 754 binary64... no checking whether C doubles are ARM mixed-endian IEEE 754 binary64... no

Accordingly 'legacy' mode isused. It is possible to check this at runtime by check if 'sys.float_repr_style' contains 'short' or 'legacy'. Calculus correctness is not garanteed with 'legacy'.

Problem is better described here:

http://stackoverflow.com/questions/29920294/what-causes-pythons-float-repr-style-to-use-legacy
https://bugs.python.org/issue7117

However, all gcc architecture use a representation compliant with IEE754. So, we can enable it unconditionnaly.

---
 package/python3/python3.mk | 7 +++++++
 1 file changed, 7 insertions(+)
package/python3/python3.mk +7 -0
index b3f31c0..cdbe6f0 100644
@@ -126,6 +126,13 @@ PYTHON3_CONF_ENV += \
 	ac_cv_working_tzset=yes \
 	ac_cv_prog_HAS_HG=/bin/false
 
+# GCC is always compliant with IEEE754
+ifeq ($(call qstrip,$(BR2_ENDIAN)),LITTLE)
+PYTHON3_CONF_ENV += ac_cv_little_endian_double=yes
+else
+PYTHON3_CONF_ENV += ac_cv_big_endian_double=yes
+endif
+
 # uClibc is known to have a broken wcsftime() implementation, so tell
 # Python 3 to fall back to strftime() instead.
 ifeq ($(BR2_TOOLCHAIN_USES_UCLIBC),y)
Signature · 1 line
2.9.3
#2 lore raw

PATCH 1/2 python: fix double format detection

In reply to Jérôme Pouiller

Hi J?r?me,

Quoted 29 lines · J?r?me Pouiller

Python is not able to detect if compiler double representation is compliant with IEE754:

checking whether C doubles are little-endian IEEE 754 binary64... no checking whether C doubles are big-endian IEEE 754 binary64... no checking whether C doubles are ARM mixed-endian IEEE 754 binary64... no

Accordingly 'legacy' mode isused. It is possible to check this at runtime by check if 'sys.float_repr_style' contains 'short' or 'legacy'. Calculus correctness is not garanteed with 'legacy'.

Problem is better described here:

http://stackoverflow.com/questions/29920294/what-causes-pythons-float-repr-style-to-use-legacy
https://bugs.python.org/issue7117

However, all gcc architecture use a representation compliant with IEE754. So, we can enable it unconditionnaly.

---
 package/python/python.mk | 7 +++++++
 1 file changed, 7 insertions(+)
package/python/python.mk +0 -0
index cc65376..0191d59 100644

package/python/python.mk PYTHON_CONF_ENV += \ · line 144

Quoted 5 lines
package/python/python.mk +2 -0
@@ -140,6 +140,13 @@ PYTHON_CONF_ENV += \
 	ac_cv_prog_HAS_HG=/bin/false \
 	ac_cv_prog_SVNVERSION=/bin/false
 
+# GCC is always complient with IEEE754
+ifeq ($(call qstrip,$(BR2_ENDIAN)),LITTLE)

I think that

  ifeq ($(BR2_ENDIAN),"LITTLE")

is more readable. That's what most other <pkg>.mk files do.

package/python/python.mk PYTHON_CONF_ENV += \ · line 148

Quoted 4 lines
package/python/python.mk +4 -0
+PYTHON_CONF_ENV += ac_cv_little_endian_double=yes
+else
+PYTHON_CONF_ENV += ac_cv_big_endian_double=yes
+endif

baruch

Signature · 3 lines
     http://baruch.siach.name/blog/                  ~. .~   Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
   - baruch at tkos.co.il - tel: +972.52.368.4656, http://www.tkos.co.il -
#3 lore raw

PATCH 1/2 python: fix double format detection

In reply to Jérôme Pouiller

Hello,

Quoted 16 lines · J?r?me Pouiller

Python is not able to detect if compiler double representation is compliant with IEE754:

checking whether C doubles are little-endian IEEE 754 binary64... no checking whether C doubles are big-endian IEEE 754 binary64... no checking whether C doubles are ARM mixed-endian IEEE 754 binary64... no

Accordingly 'legacy' mode isused. It is possible to check this at runtime by check if 'sys.float_repr_style' contains 'short' or 'legacy'. Calculus correctness is not garanteed with 'legacy'.

Problem is better described here:

http://stackoverflow.com/questions/29920294/what-causes-pythons-float-repr-style-to-use-legacy
https://bugs.python.org/issue7117

Quoted 7 lines

However, all gcc architecture use a representation compliant with IEE754. So, we can enable it unconditionnaly.

---
 package/python/python.mk | 7 +++++++
 1 file changed, 7 insertions(+)

I've applied both patches, after taking into account the comment from Baruch. Also, I verified that even ARM is not "mixed endian", by running the test case under Qemu ARM. It seems like "mixed endian" was only for the old ARM ABI, which we no longer care about.

Thanks for those patches!

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com