Re: after-rounding tininess detection and directed rounding

[John R Harrison <johnh@xxxxxxxxxxxxxxxx>]

| There is no fundamental change in the standard -- the native method for
| tininess detection can still be before or after.  What is new is the
| optional "reproducibility" attribute, which, when supported, requires
| a fixed choice for tininess detection -- currently "after rounding".

As I read it (the 1.5.0 draft of October 5, right?) the main section
on underflow, (7.5, page 48) specifies that for decimal formats
tininess is detected before rounding. By contrast, the corresponding
section (9.5, page 48) of the draft of September 17, 2006 allowed
either method.

I don't see any hedges restricting this to some "reproducible" mode,
though there is some other stuff in section 11 that discusses
"reproducible floating-point results" and then makes a further
restriction to detection *after* rounding for binary.

But I've only skimmed the new draft and not read it cover-to-cover
yet, so I may be missing some context. I hope to read the whole thing
later this week.

| Since I was confused about the precise definition of "after rounding"
| I will have to re-examine my emulation analysis.  I shall report what
| I find.

I'll be interested to know your conclusions. So far, I don't have any
strong opinions. Some code I've been writing is slightly easier with
"before rounding". But the disparity between the preferred methods of
tininess detection for binary and decimal seems a bit jarring.

John.