https://issues.dlang.org/show_bug.cgi?id=22372

Teodor Dutu <teodor.dutu@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |teodor.dutu@gmail.com
           Severity|regression                  |blocker

--- Comment #1 from Teodor Dutu <teodor.dutu@gmail.com> ---
Using run.dlang.io, I tried to find a regression, but all supported DMD versions (2.060 and newer) are printing the same incorrect output (Exception: i = 0; n = 1). My experiment can be found here: https://run.dlang.io/is/uYfTzl. This seems to be a backend issue, as both LDC and LDC-beta are working fine.

However, I came across this bug when working on this PR:
https://github.com/dlang/dmd/pull/13116. And when analysing the CI outputs, I
noticed that all 3 of DMD, LDC and GDC are failing the same test, which I've
also been able to reproduce myself:
- LDC: https://cirrus-ci.com/task/6291197929979904?logs=test_druntime#L1449
- DMD: https://cirrus-ci.com/task/5728247976558592?logs=test_druntime#L1390
- GDC: https://cirrus-ci.com/task/4883823046426624?logs=test_druntime#L1411

The failure of this test is caused by the issue I mentioned above and the code with which I reproduced the bug is based on it.

--

https://issues.dlang.org/show_bug.cgi?id=22372

Iain Buclaw <ibuclaw@gdcproject.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ibuclaw@gdcproject.org

--- Comment #2 from Iain Buclaw <ibuclaw@gdcproject.org> ---
(In reply to Teodor Dutu from comment #1)
> However, I came across this bug when working on this PR:
> https://github.com/dlang/dmd/pull/13116. And when analysing the CI outputs,
> I noticed that all 3 of DMD, LDC and GDC are failing the same test, which
> I've also been able to reproduce myself:
> - LDC: https://cirrus-ci.com/task/6291197929979904?logs=test_druntime#L1449
> - DMD: https://cirrus-ci.com/task/5728247976558592?logs=test_druntime#L1390
> - GDC: https://cirrus-ci.com/task/4883823046426624?logs=test_druntime#L1411
> 
> The failure of this test is caused by the issue I mentioned above and the code with which I reproduced the bug is based on it.
There is no bug in GDC or LDC.

$ gdc -O2 -frelease pr22372.d && ./a.out
Exception: i = 1; n = 1
$ ldc2 -O -release pr22372.d && ./pr22732
Exception: i = 1; n = 1
$ dmd -O -release pr22372.d && ./pr22732
Exception: i = 0; n = 1

--

https://issues.dlang.org/show_bug.cgi?id=22372

Imperatorn <johan_forsberg_86@hotmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |johan_forsberg_86@hotmail.c
                   |                            |om

--- Comment #3 from Imperatorn <johan_forsberg_86@hotmail.com> ---
If you make size_t i static the behaviour is correct

--

https://issues.dlang.org/show_bug.cgi?id=22372

--- Comment #4 from Teodor Dutu <teodor.dutu@gmail.com> ---
(In reply to Imperatorn from comment #3)
> If you make size_t i static the behaviour is correct

Yes, this would fix the issue, but it would be a hack rather than a solution. The bug would still persist in DMD's backend.

--

https://issues.dlang.org/show_bug.cgi?id=22372

--- Comment #5 from Imperatorn <johan_forsberg_86@hotmail.com> ---
Lol, of course. It was just an observation.

Another observation:
If you change type from size_t to int or uint the behaviour is correct

--

https://issues.dlang.org/show_bug.cgi?id=22372

Basile-z <b2.temp@gmx.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |b2.temp@gmx.com

--- Comment #6 from Basile-z <b2.temp@gmx.com> ---
Hello, I've worked on the bug a bit and found the problem location in the backend, in gloop, which performs loop optims:

`i` is optimized in the code that follows [1].

[1] https://github.com/dlang/dmd/blob/master/src/dmd/backend/gloop.d#L2966

--

https://issues.dlang.org/show_bug.cgi?id=22372

Basile-z <b2.temp@gmx.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |backend

--

https://issues.dlang.org/show_bug.cgi?id=22372

--- Comment #7 from Basile-z <b2.temp@gmx.com> ---
by digging further it looks like the problem resides in dmd.backend.gloop reference counting system. OPInd would be ignored, so when a loop body only contains an IndexExp and that the IndexExp subscript is the IV (increment value ?), DMD thinks that the IV is not used.

--

https://issues.dlang.org/show_bug.cgi?id=22372

Ketmar Dark <ketmar@ketmar.no-ip.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ketmar@ketmar.no-ip.org

--

https://issues.dlang.org/show_bug.cgi?id=22372

--- Comment #8 from Dlang Bot <dlang-bot@dlang.rocks> ---
@teodutu created dlang/druntime pull request #3585 "Use workaround until issue 22372 is fixed" mentioning this issue:

- Use workaround until issue 22372 is fixed

  Signed-off-by: Teodor Dutu <teodor.dutu@gmail.com>

https://github.com/dlang/druntime/pull/3585

--