Hey all,

After installing fpc 3.0.0, I'm getting an...

Error: function header doesn't match the previous declaration

for a (non-method) function that does exactly match its previous declaration:

function make_scale_matrix (zx:single; zy:single=NaN):t_2d_affine;

I use "NaN" extensively to indicate a null / not specified value, so it would really be a drag to lose this feature. Although the header exactly matches the previous declaration, it now objects to these NaN default parameters.

Aside from the error message being misleading/inaccurate, I didn't notice any mention of this in the user changes document; nor is this prohibition mentioned in the html documentation online, http://wiki.freepascal.org/Default_parameter

What's the status of this?

Given this feature loss, I would have to either abandon the default parameter feature and create many similar functions; or, pick another value to use (when NaN is ideal).

Thanks,
~David

It's probably due to a peciularity of the "NaN" value itself: if you ask a floating point unit to compare NaN with NaN, the returned result must always be "not equal" even though the bit patterns match exactly. To fix this, the compiler will have to compare the bit patterns of floating point default parameters rather than their values as floating point numbers.

Please file a bug report.

Jonas

I found a workaround, which is to use Infinity as the default value, instead of NaN. It's acceptable for my use, though a little awkward.

~David

David Emerson wrote

I found a workaround, which is to use Infinity as the default value, instead of NaN. It's acceptable for my use, though a little awkward.

there is another workaround.

Do not repeat the default  as
   function make_scale_matrix (zx:single; zy:single=NaN):t_2d_affine;
But implement as
   function make_scale_matrix (zx:single; zy:single):t_2d_affine;
   begin
   end;

filed

~David

Thanks, Martin! This is much better.