status of pytables across architectures

[Francesc Alted]

On Feb 15, 2012, at 6:00 PM, Yaroslav Halchenko wrote:

> 
>>>  I guess I would need to take care about those with ports teams and/or
>>>  rebuilding manually (I recall that previously that helped as a
>>>  temporary workaround)
> 
>> I found that sort of thread[1] or [2] when looking for bus-error and mips
>> it is not very clear, but at least there is some ideas in it. In fact
>> there is plenty of thoses bus-error messages...
> 
>> [1] http://postgresql.1045698.n5.nabble.com/Initdb-problem-on-debian-mips-cobalt-Bus-error-td1904368.html
>> [2] http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=463808
> 
> below are some glory details on the bus error failure (also at
> https://gist.github.com/1836685) . It looks like (and thanks to Julien Cristau
> for help with answering my silly questions) -- that src needs to be 32bit
> aligned for the casting to work correctly.  Going through the code for the
> history of src origin (defined as uint8_t*) I see no reflection of such,
> thus it eventually leads to bus error... I will leave the fix up to you,
> experts ;)
> 
> Program received signal SIGBUS, Bus error.
> blosc_d (blocksize=32768, leftoverblock=<optimized out>, src=0x132f543 "S", dest=0x1421808 "", tmp=0x1429810 "", tmp2=
>    0x1431830 "") at blosc/blosc.c:330
> 330 cbytes = sw32(((uint32_t *)(src))[0]); /* amount of compressed bytes */
[clip]

Ups, I knew this was going to happen :-/  Blosc is designed for machines that allows access to unaligned data (e.g. Intel).  Unfortunately, I'm not sure how much effort would require me to come with a version for architectures with no support for unaligned fetches (ARM?), because, among other reasons, I don't have access to one of these.

Sorry about that,

-- Francesc Alted