Re: checked int-int conversions

[Jaroslav Hajek]

On Wed, Jul 30, 2008 at 9:42 PM, John W. Eaton <address@hidden> wrote:
> On 30-Jul-2008, Jaroslav Hajek wrote:
>
> | On Wed, Jul 30, 2008 at 9:08 PM, John W. Eaton <address@hidden> wrote:
> | > On 30-Jul-2008, Jaroslav Hajek wrote:
> | >
> | > | please consider for applying the attached patch that implements
> | > | checked integer conversions - applies to converting
> | > | octave_int<T>, intXXNDArray and octave_intXX_matrix objects, as well
> | > | as Octave's builtin intXX and uintXX conversions.
> | >
> | > OK, now I'm a bit confused because I was expecting a much smaller
> | > change that would just add some warning if an attempt was made to
> | > convert a value to native int and the value was out of range.  Can you
> | > give some details about the intent of this patch?
> | >
> | > Thanks,
> | >
> | > jwe
> | >
> |
> | OK. As I have commented in the earlier thread, Octave's intX and uintX
> | builtins also do not give a warning when truncating:
> | a = int32 (2^30); b = int16 (a)
> | Neither is a warning issued if, e.g. int8_array_value () is called on
> | a octave_int32_matrix object.
> |
> | I've added a static flag to the octave_int<T> class that gets enabled
> | when data are truncated by the conversion constructor. The
> | intY_array_value methods of octave_intX_matrix classes check the flag
> | after the conversion and issue a warning when truncation occured.
> | I've also modified the conversion operators in ops.h to ask the
> | octave_value object directly for the required type instead of doing
> | the conversion themselves. This makes the intX and uintX builtins warn
> | if truncation is done:
> |
> | a = int32 (2^30); b = int16 (a)
> | warning: data truncated converting from int32 to int16
> | 32767
> |
> | a warning is also issued if negative intX is converted to uintX.
>
> OK, now I also see that Matlab has a similar optional warning.  I'll
> apply the patch.
>
> In Matlab, the warning is controlled with the warning ID
> MATLAB:intConvertOverflow.  Should we use the same tag (but with an
> Octave: prefix)? Then if we also modified the warning_with_id
> function to convert a "MATLAB:" warning ID prefix to "Octave:" before
> comparing IDs it would be easier to write portable code.
>

A good idea - I forgot to check the Matlab counterparts. Should I
resubmit the patch (given that this is, as I understand, a trivial
change)?

> Thanks,
>
> jwe
>

-- 
RNDr. Jaroslav Hajek
computing expert
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz