Experimental Physics and
Industrial Control System

Matt Newville <[email protected]> · Wed, 22 Sep 2010 13:52:20 -0500

Hi Andrew,

Thanks for all the information. I'll try to add tests for these cases.

--Matt

On Wed, Sep 22, 2010 at 10:37 AM, Andrew Johnson <[email protected]> wrote:
> Hi Matt,
>
> On Wednesday 22 September 2010 09:56:55 Matt Newville wrote:
>> OK, I was able to reproduce the problem on a linux x86_64 machine.
>> For Python, I see that
>>
>> >>> import ctypes
>> >>> ctypes.c_long == ctypes.c_int
>>
>> is True on 32 bit systems but is False on 64-bit linux.  As I said, I
>> was mapping Epcics DBR_LONG to a ctypes.c_long.   My suspicion is that
>> an Epics DBR_LONG is actually not a "long" on 64-bit machines, but a
>> 32-bit integer, which is to say an int.  Perhaps someone can confirm,
>> correct, or clarify this.
>
> That is correct, there was too much EPICS software that assumed a DBR_LONG was
> always 32 bits, so we map DBR_LONG (and DBF_LONG inside the IOC) to an
> epicsInt32 which is our architecture-independent typedef.  If someone needs
> anything longer I'd suggest using a DBR_DOUBLE, which can accurately represent
> all integers up to 52 bits long (above that you can find that x+1 == x).
>
>> I use ctypes.c_long and ctypes.c_ulong all over dbr.py for the mapping
>> of various CA structures, including event and connection handling and
>> in the more complex data (DBR_CTRL_XXX, etc).  But callbacks and at
>> least some complex data types appear to be working, so some portions
>> of these structures must really be longs.  I'll have to do more
>> testing on 64-bit machines to better sample over all the mappings of
>> the C structures.
>
> Things might appear to work on little-endian machines but not on big-endian or
> vice versa; test this on both (Sparc or PowerPC) and x86 families if you can.
> The DBR_XXX_LONG structures do use 32-bit values for the limit values.
>
> On a related subject, be aware that in a DBR_CTRL_DOUBLE the alarm limits will
> be NaN values if the relevent xxSV field is NO_ALARM; you might want to make
> sure that Python handles them properly.
>
> - Andrew
> --
> The best FOSS code is written to be read by other humans -- Harald Welte

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System