Bug 8624 – Regression: CTFE: long comparisons completely broken

dmd 2.060 x64 linux. code: int generateHashAndValueArrays() { import std.string; assert(H("a") != H("A"), xformat("%x == %x", H("a"), H("A"))); return 1; } enum s = (generateHashAndValueArrays()); ulong H(string str) { if(str[0] == 'a') return 0x6161636772; return 0x4161636772; } fireworks: HtmlEntities.d(4): Error: ['6','1','6','1','6','3','6','7','7','2',' ','=','=',' ','4','1','6','1','6','3','6','7','7','2'][0LU..24LU] ehh, HtmlEntities.d(4): Error: "6161636772 == 4161636772"

This is one of the worst bugs of all time. I can't believe this slipped through the test suite. The top 32 bits of longs is entirely ignored in CTFE == and != comparisons. Reduced test case: ==================== int bug8624() { long m = 0x1_0000_0000; assert(m != 0); return 1; } static assert( bug8624() ); ==================== It's a typo in interpret.c, which results in an implicit conversion from long to int. This might be worth an emergency release.

In which version did it last work correctly?

(In reply to comment #1) > It's a typo in interpret.c, which results in an implicit conversion from long > to int. > This might be worth an emergency release. Please post pull request, and I'll get it pulled at least.

Commits pushed to master at https://github.com/D-Programming-Language/dmd https://github.com/D-Programming-Language/dmd/commit/9a732a03c9f671a7c45a9f7dbe485a0b3489f997 Fix issue 8624 Regression: CTFE: long comparisons completely broken An implicit narrowing cast from long to int. Catastrophic. https://github.com/D-Programming-Language/dmd/commit/475178efa63597d6d55a074406678ab92f1075ed Merge pull request #1112 from donc/regression8624ctfe Fix issue 8624 Regression: CTFE: long comparisons completely broken