dmd -O produces incorrect code?

[1100110]

On 02/10/2013 09:26 PM, 1100110 wrote:
> On 02/10/2013 07:14 PM, 1100110 wrote:
>> I've taken std.mmfile, and been playing with it.
>>
>> I won't lie, mostly I've just been messing with it.
>> The idea was originally to fix opDollar for mmfile (which I can push if
>> it will be accepted.)
>>
>> I turned it into a template for funsies.
>> And a struct. For funsies.
>>
>> And it Works! mostly.
>>
>> But only if you don't compile with -O.
>> -release works, -inline works, -noboundscheck works.
>>
>> heres an example:
>> ---
>> import std.conv, std.stdio;
>> void main()
>> {
>> auto mm = mmFile!string("mmTest.d");
>> string mt = mm[0..50];
>> string ms = mt[0..50];
>> string ss = mm.data;
>>
>> writeln(mm[0..$]);//prints the entire file.
>> writeln(ms);//prints the first 50 characters.
>>
>> writeln(mm.data.ptr);//data is the internal string.
>> writeln(ms.ptr);//all of these should be equal.
>> writeln(mt.ptr);//But this one and the one above are not.
>> writeln(ss.ptr);//also nothing gets printed with -O, it segfaults.
>>
>> return;
>> }
>> --
>>
>> Here is an example output with rdmd mmTest.d:
>> //the above file[0..$]
>> //the about file[0..50]
>> 7F47535F5000
>> 7F47535F5000
>> 7F47535F5000
>> 7F47535F5000
>>
>> Here is outut with rdmd -O mmTest.d:
>> �
>> �
>> 7FEA61B77000
>> 406470
>> 406470
>> 7FEA61B77000
>>
>> What's up with that?
>
> Found it! I was destroy(class) manually allocated class.
> That was my fault. Completely. I knew I would be embarrassed if I posted
> this. =P
>
> I am now in love with cgdb.
> Go get it. It's beautiful.

Well, mostly. It still has different output on -O, but no longer crashes.