I'm seeing what appears to be a bug with the -O flag in dmd. Here is a reduced test case:

struct SomeStruct
{
    long value;
}
bool isNumberOne(int i)
{
	SomeStruct l;
	if(i == 1)
		l = SomeStruct(10);
	return (l == SomeStruct(10));
}
void main()
{
    if (!isNumberOne(1))
        assert(false);
}

See: https://run.dlang.io/is/zlrDvy

rdmd test.d    # Runs fine
rdmd -O test.d # Assert trips

The assert trips when run with "rdmd -O test.d" and it does not trip when run with "rdmd test.d".

This bug is observed in all compiler versions from dmd-2.060 through to dmd-2.095! Please shed some light on the issue – any help would be greatly appreciated.

Thanks,
Saurabh

On Mon, Jan 25, 2021 at 06:07:46PM +0000, Saurabh Das via Digitalmars-d-learn wrote:
> I'm seeing what appears to be a bug with the -O flag in dmd.
[...]
> The assert trips when run with "rdmd -O test.d" and it does not trip when run with "rdmd test.d".

It's probably a bug. File a bug on bugzilla: https://issues.dlang.org


> This bug is observed in all compiler versions from dmd-2.060 through to dmd-2.095! Please shed some light on the issue – any help would be greatly appreciated.
[...]

DMD's backend is known to have obscure bugs that crop up every so often. If this issue is being a showstopper, consider switching to LDC or GDC instead.  (You might want to try the same code with LDC/GDC anyway, just to see whether they have the same problem. My guess is they don't.)


T

-- 
People demand freedom of speech to make up for the freedom of thought which they avoid. -- Soren Aabye Kierkegaard (1813-1855)

On Monday, 25 January 2021 at 18:19:24 UTC, H. S. Teoh wrote:
> On Mon, Jan 25, 2021 at 06:07:46PM +0000, Saurabh Das via Digitalmars-d-learn wrote:
>> [...]
> [...]
>> [...]
>
> It's probably a bug. File a bug on bugzilla: https://issues.dlang.org
>
>
>> [...]
> [...]
>
> DMD's backend is known to have obscure bugs that crop up every so often. If this issue is being a showstopper, consider switching to LDC or GDC instead.  (You might want to try the same code with LDC/GDC anyway, just to see whether they have the same problem. My guess is they don't.)
>
>
> T

LDMD2 seems to not suffer from this issue.

On Mon, Jan 25, 2021 at 06:45:43PM +0000, Saurabh Das via Digitalmars-d-learn wrote:
> On Monday, 25 January 2021 at 18:19:24 UTC, H. S. Teoh wrote:
[...]
> > It's probably a bug. File a bug on bugzilla: https://issues.dlang.org
[...]
> > DMD's backend is known to have obscure bugs that crop up every so often.  If this issue is being a showstopper, consider switching to LDC or GDC instead.  (You might want to try the same code with LDC/GDC anyway, just to see whether they have the same problem. My guess is they don't.)
[...]
> LDMD2 seems to not suffer from this issue.

No surprise there. I recommend using LDC (ldmd2) instead.

Still file the bug, though, so that somebody will fix it eventually.


T

-- 
There is no gravity. The earth sucks.

On Monday, 25 January 2021 at 19:18:47 UTC, H. S. Teoh wrote:
> On Mon, Jan 25, 2021 at 06:45:43PM +0000, Saurabh Das via Digitalmars-d-learn wrote:
>> On Monday, 25 January 2021 at 18:19:24 UTC, H. S. Teoh wrote:
> [...]
>> > It's probably a bug. File a bug on bugzilla: https://issues.dlang.org
> [...]
>> > DMD's backend is known to have obscure bugs that crop up every so often.  If this issue is being a showstopper, consider switching to LDC or GDC instead.  (You might want to try the same code with LDC/GDC anyway, just to see whether they have the same problem. My guess is they don't.)
> [...]
>> LDMD2 seems to not suffer from this issue.
>
> No surprise there. I recommend using LDC (ldmd2) instead.
>
> Still file the bug, though, so that somebody will fix it eventually.
>
>
> T

Thanks for the inputs. I'll fill a bug report about it.