Bug 3835 – ref foreach does not work in CTFE

This code (that uses CTFE to do something that can be useful, it initializes an array with values compiled at compile time) compiles and works correctly, but I think it's too much fragile: if you change the for loop a little, or other details in genFactorials(), the program stops compiling: int factorial(int n) { if (n == 0) return 1; return n * factorial(n - 1); } int[] genFactorials(int n) { int[] result; for (; n >= 0; --n) result = factorial(n) ~ result; return result; } const int N = 13; static const auto factorials = cast(int[N])genFactorials(N - 1); void main() {} If you need them, I can show many variants of this program that don't compile/work. I think to be useful in real programs CTFE has to be less fragile than this.

(In reply to comment #0) > This code (that uses CTFE to do something that can be useful, it initializes an > array with values compiled at compile time) compiles and works correctly, but I > think it's too much fragile: if you change the for loop a little, or other > details in genFactorials(), the program stops compiling: Please put in an example that it wouldn't compile!

> Please put in an example that it wouldn't compile! OK. This two versions of genFactorials print the wrong results: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 import std.stdio: writeln; int factorial(int n) { if (n == 0) return 1; return n * factorial(n - 1); } int[] genFactorialsA(int n) { int[] result = new int[n + 1]; foreach (i, ref el; result) el = factorial(i); return result; } int[N] genFactorialsB(int N)() { int[N] result; foreach (i, ref el; result) el = factorial(i); return result; } enum int N = 13; static enum auto factorials1 = cast(int[N])genFactorialsA(N - 1); enum auto factorials2 = genFactorialsB!13; void main() { writeln(factorials1); writeln(factorials2); } This version doesn't compile, the compiler raises: test7b.d(7): Error: pure function 'factorial' cannot call impure function 'factorial' test7b.d(12): Error: pure function 'genFactorials' cannot call impure function 'factorial' test7b.d(17): Error: cannot evaluate genFactorials(12) at compile time test7b.d(17): Error: cannot evaluate genFactorials(12) at compile time import std.stdio: writeln; pure int[] genFactorials(int n) { pure static int factorial(int n) { if (n == 0) return 1; return n * factorial(n - 1); } int[] result = new int[n + 1]; foreach (i; 0 .. n+1) result[i] = factorial(i); return result; } enum int N = 13; static enum auto factorials = cast(int[N])genFactorials(N - 1); void main() { writeln(factorials); }

Looks like foreach do not work in CTFE. Here is example with two equivalent functions, one uses 'for', other 'foreach' loops: uint[256] crc32_table_for() { uint[256] table; for (uint i = 0; i < 256; ++i) { uint crc = i; for (int j = 0; j < 8; j++) crc = crc & 1 ? (crc >> 1) ^ 0xEDB88320 : crc >> 1; table[i] = crc; } return table; } uint[256] crc32_table_foreach() { uint[256] table; foreach (n, ref i; table) { uint crc = n; foreach (j; 0 .. 8) crc = crc & 1 ? (crc >> 1) ^ 0xEDB88320 : crc >> 1; i = crc; } return table; } immutable crc32_table1 = crc32_table_for(); immutable crc32_table2 = crc32_table_foreach(); import std.stdio; void main() { writefln("%s", crc32_table1); writefln("%s", crc32_table2); assert (crc32_table1 == crc32_table2); }

I think foreach(x; items) works in CTFE, but foreach(ref x; items) doesn't work well.

The first two cases are already present in bug 2411. I don't know about the third.

I created bug 3912 for the third bug mentioned in the comments, which has nothing to do with CTFE. Maybe this bug can now be marked as a duplicate of bug 2411?

You can mark this as duplicate. But when an improvement is created, it's necessary to try it with *all* duplicated examples too, because you must be sure they are really problems with the same cause.

Bug 2411 is NOT the same as this. (2411 does not involve CTFE). Changing title.

In dmd 2.043 my second tests cases works: import std.stdio: writeln; pure int[] genFactorials(int n) { pure static int factorial(int n) { if (n == 0) return 1; return n * factorial(n - 1); } int[] result = new int[n + 1]; foreach (i; 0 .. n+1) result[i] = factorial(i); return result; } enum int N = 13; static enum auto factorials = cast(int[N])genFactorials(N - 1); void main() { writeln(factorials); } The first test case doesn't work yet.

A reduced test case: int foo() { int[1] arr; foreach (ref el; arr) el = 10; return arr[0]; } enum int r = foo(); void main() { assert(r == 10); }

This bug is worse than it might seem from reading the »rejects-valid« keyword – DMD 2.051 happily executes ref foreach during compile-time, but generates wrong results silently. Should we add a »wrong-ctfe« tag or something like that?

(In reply to comment #11) > This bug is worse than it might seem from reading the »rejects-valid« keyword – > DMD 2.051 happily executes ref foreach during compile-time, but generates wrong > results silently. > > Should we add a »wrong-ctfe« tag or something like that? No, incorrect CTFE should just be marked as wrong-code. Yup, this one's important.

Fixed. D2: https://github.com/D-Programming-Language/dmd/commit/0219a5f0dcc88076759a4c472582954d0fb804b0