Bug 4603 – array(iota(1, 0)) error

This generates an 'Access Violation' with dmd 2.048beta: import std.range; void main() { array(iota(1, 0)); } The expected result of this array() is an empty dynamic array of signed integers.

If you write --- import std.range; import std.stdio; void main() { foreach (a; iota(1, 0)) { writeln(a); } } --- the program will never stop writing. Looks like an integer overflow bug.

Created attachment 707 Patch for range.d to produce an empty range when pastLast <= current. This patch produces an empty range when pastLast <= current. But this is inconsistent with the behavior of floating point iota ranges.

Created attachment 708 Patch for range.d to throw when pastLast <= current. This patch simply enforces pastLast > current, so it is consistent with the floating point iota range. But it may be less useful.

Fixed in changeset 1901. I chose to throw in this case. The proper way to create an empty range would be to do iota(0, 0), or iota(1, 1). The main reason for this choice was consistency with array slicing. For example, if bounds checking is enabled, array[3..2] throws a RangeError.

Throwing? No, look at Python: >>> range(1, 0) [] It needs to return an empty range. This: foreach (i; iota(1, 0)) is the same as: foreach (i; 1 .. 0) Or: for (int i = 1; i < 0; i++) It's like an empty loop.

And by the way, in Python This produces an empty string slice: >>> "abcdefg"[3:2] ''

Thank you for improving this bug. But I reopen the bug because I think raising an error is the wrong design. As I have explained iota() is to be meant as the ranges in for/foreach loops, so iota(1, 0) is like an empty loop, it's not an error.

See also bug 6222

This now seems to work as you want in DMD 2.059. The following: import std.stdio, std.range; void main() { auto empty_array = array(iota(1, 0)); writeln("type: ", typeof(empty_array).stringof, ", empty? ", empty_array.empty, ", contents: ", empty_array); } Outputs: type: int[], empty? true, contents: [] I'd say it can be closed and marked RESOLVED now.