code review 5577068: spec: emphasize that compilers may round or reject cons...

spec: emphasize that compilers may round or reject constants.

Fixes issue 2789.

[remyoudompheng, 2012-01-28 11:18:06]

Hello golang-dev@googlegroups.com (cc: golang-dev@googlegroups.com,
remy@archlinux.org),

I'd like you to review this change to
https://go.googlecode.com/hg/

[remyoudompheng, 2012-01-28 11:20:21]

This proposal aims at clearly making issue 2789 a non-issue and explains why it
is legal to have

package main

func main() {
	println(0.1 + 0.2 == 0.3)
	println(int64(1e2012 + 1 - 1e2012))
}

print:

false
0

[bsiegert, 2012-01-28 12:15:20]

http://codereview.appspot.com/5577068/diff/3001/doc/go_spec.html
File doc/go_spec.html (right):

http://codereview.appspot.com/5577068/diff/3001/doc/go_spec.html#newcode597
doc/go_spec.html:597: In particular, a compilar may refuse to handle integral
numeric constants that
compiler

[remyoudompheng, 2012-01-28 13:04:53]

Hello golang-dev@googlegroups.com, bsiegert@gmail.com (cc:
golang-dev@googlegroups.com, remy@archlinux.org),

Please take another look.

[rsc, 2012-01-29 16:02:00]

Please add
    Fixes issue 2789.
to the CL description.

[r, 2012-01-29 16:07:15]

i agree this should be settled but we need to define it better

http://codereview.appspot.com/5577068/diff/5001/doc/go_spec.html
File doc/go_spec.html (right):

http://codereview.appspot.com/5577068/diff/5001/doc/go_spec.html#newcode558
doc/go_spec.html:558: Precision of numeric constants may be subject to
implementation
s/Precision/The precision/

however, this feels too vague to me. i want to write code that i know will work;
what can i depend on?

http://codereview.appspot.com/5577068/diff/5001/doc/go_spec.html#newcode599
doc/go_spec.html:599: floating-point or complex constants.
again, this is too vague

[remyoudompheng, 2012-01-29 16:35:44]

http://codereview.appspot.com/5577068/diff/5001/doc/go_spec.html
File doc/go_spec.html (right):

http://codereview.appspot.com/5577068/diff/5001/doc/go_spec.html#newcode558
doc/go_spec.html:558: Precision of numeric constants may be subject to
implementation
What about moving the "Implementation restrictions" paragraph up here?

http://codereview.appspot.com/5577068/diff/5001/doc/go_spec.html#newcode599
doc/go_spec.html:599: floating-point or complex constants.
Actually, what does the "any machine type" thing means. Is it about hardware
support? Since the spec already defines what are the integer and floating-point
types, can we say 128 bits here?

We could rewrite the paragraph as "A compiler must guarantee exact constant
arithmetic for all integral values under 2^128. Rounding errors for operations
on floating-point or complex constants must be not greater than those obtained
using a mantissa and exponent twice as large as float64."

[remyoudompheng, 2012-02-10 07:45:36]

*** Abandoned ***