[Buildroot] [PATCH v4 0/2] Micropython package

[Vicente Olivert Riera]

Dear Chris Packham,

On 09/18/2015 09:22 AM, Chris Packham wrote:
> On Fri, Sep 18, 2015 at 7:32 PM, Thomas Petazzoni
> <thomas.petazzoni at free-electrons.com> wrote:
>> Chris,
>>
>> On Thu, 17 Sep 2015 11:29:57 +1200, Chris Packham wrote:
>>
>>> Chris Packham (2):
>>>   micropython: new package
>>>   micropython-lib: new packages
>>
>> There are a number of build failures caused by Micropython in the
>> autobuilders:
>>
>>  * On MIPS64
>>    http://autobuild.buildroot.org/results/99a/99a610c72877fef53d93e1a923bd5c71b3376618/build-end.log
>>
>>  * On Xtensa
>>    http://autobuild.buildroot.org/results/670/670907760d6b3e7eb46b433647b392cec3bafdcd/build-end.log
>>
>>  * On SH4A
>>    http://autobuild.buildroot.org/results/f5a/f5a2c563b8bdab9e345960a5b92d31f3f92dba00/build-end.log
>>
>> See http://autobuild.buildroot.org/?reason=micropython-v1.4.5 for the
>> complete list of build failures.
>>
>> If you could look into these, it would be helpful.
> 
> Xtensa and SH4A look like they need MICROPY_GCREGS_SETJMP I'll send a
> patch for them shortly. The MIPS64 one looks different so might need
> some investigation.

as the build error says, a right shift is being performed in the
py/objint_mpz.c file, and the offending one is at line #54:

MP_SSIZE_MAX >> MPZ_DIG_SIZE * 4

in MIPS64 n64 the MP_SSIZE_MAX is 64-bit wide and the value of
MPZ_DIG_SIZE 16, so MPZ_DIG_SIZE * 4 is 64. That shift causes a warning
because its trying to shift 64 bits of a 64-bit variable:

py/objint_mpz.c:54:5: error: right shift count >= width of type [-Werror]

MIPS32 and MIPS64 n32 are fine because MP_SSIZE_MAX is 32-bit wide and
the #if clauses in the py/objint_mpz.c protect that to happen.

Look at this test for MIPS32:

$ mips-linux-gnu-gcc -mips32r2 -EB ssize_t.c -o ssize_t
$ qemu-mips ssize_t
MP_SSIZE_MAX: 32 bits
MPZ_DIG_SIZE: 16
right-shifting 0 bits
right-shifting 16 bits

And now look at the same test for MIPS64 n64:

$ mips-linux-gnu-gcc -mips64r2 -mabi=64 -EB ssize_t.c -o ssize_t
ssize_t.c: In function 'main':
ssize_t.c:24:2: warning: right shift count >= width of type
  (MP_SSIZE_MAX >> MPZ_DIG_SIZE * 4) & DIG_MASK;
  ^
$ qemu-mips64 ssize_t
MP_SSIZE_MAX: 64 bits
MPZ_DIG_SIZE: 16
right-shifting 0 bits
right-shifting 16 bits
right-shifting 32 bits
right-shifting 48 bits
right-shifting 64 bits

I don't know what's the purpose of the code in py/objint_mpz.c and I
don't know if they really want to do a 64-bit shift of a 64-bit
variable. But the thing is that since it's compiled with the -Werror
flag, that warning makes the compilation to fail.

If they really want to do such a shift, then they should remove the
-Werror. But if they don't want to do such a shift, then more #if
clauses are needed in order to prevent that to happen. Perhaps you could
talk with upstream about this.

Regards,

Vincent.

> 
>>
>> Thanks!
>>
>> Thomas
>> --
>> Thomas Petazzoni, CTO, Free Electrons
>> Embedded Linux, Kernel and Android engineering
>> http://free-electrons.com
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
>