[L2Ork-dev] PD_BIGORSMALL

[Jonathan Wilkes]

Hi list,

Ok, I'm delving into IEEE 754 a bit more, and I want to
get a handle on the m_pd.h public interface changes before
delving too much deeper into the double-precision project.

Here's our recent code change that Pranay ported from
pd-double to check for subnormals (a.k.a. denormals), nans,
and some other desiderata:

https://git.purrdata.net/jwilkes/purr-data/commit/d0d6b54deb1b803c67d85bd5d123d3da98faf495

So basically we've replaced the old direct type-punning code
with a union. From what I can tell, there are two benefits to
the new approach:

* we can error out on strict aliasing violations
and thus ensure that no such violations accidentally get
added in *other* parts of the code (where, unlike this macro,
they likely would be bugs).

* we can just do a comparison on the 2nd array element
for double-precision

However, we could still get silently borked for the following
three cases:

* systems where sizeof(int) != sizeof(float) for single-precision Purr Data

* systems where sizeof(int) * 2 != sizeof(double) for double-precision Purr
Data

* big endian CPUs

Should we add a preprocessor check for the type sizes and
error out if they don't meet our assumptions?

As far as big endian, I don't see anything in the way of
consumer CPUs (and RISCV is apparently littleendian).

Finally-- it appears that there is no issubnormal in
math.h. Anyone have a clue why that is?

-Jonathan