[libstdc++] numeric_limits: round_style, round_error

[Gabriel Dos Reis]

Mark Mitchell <mark@codesourcery.com> writes:

| > Next, round_error is intimately tied to round_style.  When
| > round_style is round_to_nearest, this value should be 0.5.
| > While this is a function, not a constant, I think it would
| > be confusing to vary at runtime when round_style doesn't.
| 
| A technical comment, which I'm not sure applies, since I didn't try to
| read through all of the patches.
| 
|   If you have changed the value of *any* constant exposed by a V3
|   header, you have changed the ABI.

I'm not sure that really applies.  I don't think those values are part
of the ABI -- before this patch, they were practically useless. 

[...]

| *I* don't, for things like this; if we're fixing bugs in the library,
| that's good.

FWIW, those are bugs.  It all started exactly with fixing such bugs in
this area -- and I'm grateful to RTH for having spent time reviewing
my patches and implementing better alternatives.

| However, we should point out in case somebody cares.

I understand.

-- Gaby