Bug 5191 – Combination of pure and nothrow result in a function that does nothing

import std.stdio; struct Foo { void add(T)(T value) pure nothrow { this.value += value; } this(int value) { this.value = value; } int value; } void main() { auto foo = Foo(5); assert(foo.value == 5); foo.add(2); writeln(foo.value); assert(foo.value == 7); foo.add(3); writeln(foo.value); assert(foo.value == 10); foo.add(3); writeln(foo.value); assert(foo.value == 13); } If you remove either the pure _or_ the nothrow (or both) from add(), you get this output: 7 10 13 However, if add() is both pure and nothrow, then you get an assertion failure: 5 core.exception.AssertError@test(25): Assertion failure ---------------- ./test(onAssertError+0x2e) [0x808d40e] ./test(_d_assertm+0x16) [0x8085536] ./test(void test.__assert(int)) [0x80828d6] ./test(_Dmain+0x43) [0x807f7a7] ./test(extern (C) int rt.dmain2.main(int, char**)) [0x8085746] ./test(extern (C) int rt.dmain2.main(int, char**)) [0x80856a0] ./test(extern (C) int rt.dmain2.main(int, char**)) [0x808578a] ./test(extern (C) int rt.dmain2.main(int, char**)) [0x80856a0] ./test(main+0x96) [0x8085646] /usr/lib32/libc.so.6(__libc_start_main+0xe6) [0xf75d7c76] ./test() [0x807f6a1] For some reason, the combination of pure and nothrow results in the function being a no-op. Personally, since I try to make as many functions nothrow and as many functions pure as possible, this is a serious fly in the ointment of both pure and nothrow.

Nasty. It's because pure member functions are internally being marked as const somehow. That needs to be fixed. In DMD2.049, this code didn't compile.

Extra test case (add to the end of the original test case): --- void delegate (int) pure nothrow dg = &foo.add!(int); dg(7); assert(foo.value == 20); -- PATCH: e2ir.c, callfunc(), line 294. To determine if a function is strongly pure, you need to look at the function declaration, not just the function type, since the 'this' parameter isn't part of the type. ----- else if (ep) { /* Do not do "no side effect" calls if a hidden parameter is passed, * as the return value is stored through the hidden parameter, which * is a side effect. */ - e = el_bin((tf->purity == PUREstrong && tf->isnothrow && (retmethod != RETstack)) ? + e = el_bin(((fd ? fd->isPure() : tf->purity) == PUREstrong && tf->isnothrow && (retmethod != RETstack)) ? OPcallns : OPcall,tyret,ec,ep); if (tf->varargs) e->Eflags |= EFLAGS_variadic; } else - { e = el_una((tf->purity == PUREstrong && tf->isnothrow && (retmethod != RETstack)) ? + { e = el_una(((fd ? fd->isPure() : tf->purity) == PUREstrong && tf->isnothrow && (retmethod != RETstack)) ? OPucallns : OPucall,tyret,ec);

*** Issue 5277 has been marked as a duplicate of this issue. ***

Note that inlining must be turned off for this test case to fail.