fix Issue 6531 - assertion failure in std.range.iota #1673
Conversation
| } | ||
| else | ||
| { | ||
| if (pastEnd <= nextDown(end)) ++count; |
There was a problem hiding this comment.
If end is a whole number, then end is nextDown(end) ... correct? http://dlang.org/phobos/std_math.html#.nextDown documents everything but the normal behavior :/
In any case, doesn't that mean the if conditions aren't strictly equivalent in run-time and ctfe? (< vs <=)
There was a problem hiding this comment.
Wait, never mind, I think I got it.
There was a problem hiding this comment.
Nope. Now I don't even understand why this code is here at all. How can pastEnd < end at all? Doesn't this require wrap around, which can only happen with integrals? This is confusing :/
There was a problem hiding this comment.
This code handles the case where count is rounded incorrectly (count = to!size_t((end - beg) / step))).
There was a problem hiding this comment.
Ah... I miss-understood what "nextDown" does. Makes sense now.
Wouldn't:
real fcount = trunc((end - beg) / step);
count = to!size_t((fcount));give an exact count, without further need for any rounding correction? It's real from end to end...?
- The assertion fails because of excess precision. - Can't use nextUp/nextDown in ctfe, but currently ctfe always computes with maximum precision so it doesn't run into this issue.
|
Pinging @MartinNowak : Did you read my last message? I think your pull is correct. It is just producing "unexpected" (but not incorrect) results. Fix it up? |
| assert(iota(0.0, 3.0, 0.03).length == 100); | ||
| assert(iota(0.0, 3.0, 0.03).back < 3.0); | ||
| assert(iota(-936596.062f, 373124.094f, 327430.031f).length == 5); | ||
| assert(iota(0.0, 3.6, 0.03).length == 120, text(iota(0.0, 3.6, 0.03).length)); |
There was a problem hiding this comment.
Test failure: This is just missing an import to std.conv.
|
I'll close this until I have time to verify the solution. Or maybe somebody else fixes the issue. |
always computes with maximum precision so it doesn't
run into this issue.
https://d.puremagic.com/issues/show_bug.cgi?id=6531