Bug 6098 – Static array corruption of data

import std.math; import std.stdio; import std.range; auto makeSine(int TableSize)() { double[TableSize] result; // double[] result = new double[](TableSize); // this fixes the issue foreach (index, ref sample; result) { sample = cast(double)sin((cast(double)index / cast(double)TableSize) * PI * 2.); } writeln(result); return cycle(result); } void main() { auto sineTable = makeSine!(10); writeln(take(sineTable, 10)); writeln(take(sineTable, 10)); double[] arr; foreach (sample; take(sineTable, 10)) { arr ~= sample; } writeln(arr); } Results: [0, 0.587785, 0.951057, 0.951057, 0.587785, -5.42101e-20, -0.587785, -0.951057, -0.951057, -0.587785] [6.14906e-318, 2.59842e-307, 2.31044e-317, 0.951057, 0.587785, -5.42101e-20, 2.64149e-308, 2.08636e-317, 1.82805e-307, 2.641e-308] [6.14906e-318, 2.59842e-307, 2.31044e-317, 2.59842e-307, 2.31044e-317, -5.42101e-20, 2.64149e-308, 2.08636e-317, 1.82805e-307, 2.641e-308] [6.14906e-318, 2.64128e-308, 2.49803e-307, 2.2817e-314, 6.3666e-314, -5.42101e-20, 2.64149e-308, -nan, 1.82805e-307, 2.64126e-308] If bugzilla makes it hard to read I've pasted the results here: http://paste.pocoo.org/show/400027/

Actually this has nothing to do with CTFE: import std.math; import std.stdio; import std.range; auto makeSine() { double[10] result; //~ double[] result = new double[](TableSize); // this fixes the issue foreach (index, ref sample; result) { sample = cast(double)sin((cast(double)index / cast(double)10) * PI * 2.); } writeln(result); return cycle(result); } void main() { auto sineTable = makeSine(); writeln(take(sineTable, 10)); }

Actually, it's expected :) result is on the stack, which goes away when you return from the function (this is why the heap-allocating version has no issues). I'm assuming the whole shenanigans with cycle and take are so the compiler doesn't complain? That should probably be a hint...

Smiley face, but this is a subtle thing without any warnings from the compiler. I'm using cycle here for good purpose and not to drown the compiler, take would be used to fill a buffer of some certain length. I'll just use the heap, it's not a problem. I just assumed using a range on stack-allocated data would copy the data when returning the range from the function.

Actually I can just dup manually instead of using the heap: return cycle(result.dup);

(In reply to comment #3) > I'm using cycle here for good purpose and not to drown the compiler, take would > be used to fill a buffer of some certain length. OK, it just seemed strange that you were cycling something of length 10, and then only taking 10 elements from it. I wondered what the purpose of this was. Sorry for the assumption.

Here's a more complete example: http://paste.pocoo.org/show/400066/ Basically it's a sinewave that repeats forever. Changing the stepping will increase the pitch, the buffer length depends on the soundcard (for example the soundcard might require a couple of buffers of a certain length to be filled each second), table size basically increases the resolution or smoothness of the curve. I think that's it, anyway. I'm still new to the whole concept of audio programming. :)

(In reply to comment #4) > Actually I can just dup manually instead of using the heap: > > return cycle(result.dup); .dup creates a copy on the heap. This is a safer implementation: import std.stdio, std.math, std.algorithm, std.range; auto makeSine() { return cycle(map!q{ sin(a * PI) }(iota(0, 2, 0.2))); } void main() { auto sineTable = makeSine(); writeln(take(sineTable, 10)); } Regarding the unsafety of the version that uses a fixed sized array, I'd like the compiler to warn better against such bugs.

(In reply to comment #7) > (In reply to comment #4) > > Actually I can just dup manually instead of using the heap: > > > > return cycle(result.dup); > > .dup creates a copy on the heap. > > > This is a safer implementation: > > import std.stdio, std.math, std.algorithm, std.range; > > auto makeSine() { > return cycle(map!q{ sin(a * PI) }(iota(0, 2, 0.2))); > } > > void main() { > auto sineTable = makeSine(); > writeln(take(sineTable, 10)); > } > > > Regarding the unsafety of the version that uses a fixed sized array, I'd like > the compiler to warn better against such bugs. This will evaluate 'sin(a * PI)' N times, which totally goes against the reason of making a sine table. You should 'array' the 'map' before 'cycle'-ing.