Bug 14551 – scope(failure) block causes bloat in nothrow functions

Consider the example shown in http://goo.gl/EY3o7p. There, simply adding an empty scope(failure) block causes an increase in generated code. However, the function is statically known to not throw so the entire scope(failure) statement can be eliminated in the front-end.

http://asm.dlang.org/ looks to be down. Would be nice to know actual example. At least in gdc, I did found that `scope(failure)` in `nothrow` functions (or even without manually indicating them as such) compiles to zero code, and is eliminated to nothing. However it doesn't warn about dead code, which IMHO it should. Example: ``` int i = 0; size_t f(string x) nothrow { scope(failure) i = 1; return 2 + x.length * x.length; } int main(string[] args) { return cast(int)f(args[1]); } ``` Compiles with `gdc-9 -fno-bounds-check -O2 -Wall` to: ``` 0000000000001230 <_Dmain>: 1230: 48 8b 46 10 mov 0x10(%rsi),%rax 1234: 48 0f af c0 imul %rax,%rax 1238: 48 83 c0 02 add $0x2,%rax 123c: c3 retq 123d: 0f 1f 00 nopl (%rax) ``` Wow.

Also it looks to be possibly similar issue to https://issues.dlang.org/show_bug.cgi?id=13575

The original code is not reachable anymore so we cannot fix this without a test case.