I've attached a short example (mostly declarations, only a few executable statements) for which the 13.0 compiler is generating bad object code. The offending statement is an intrinsic assignment of a derived type with pointer component where the rhs is a user-defined constructor. I point out a couple variations where the code will run correctly.
| Вложение | Размер |
|---|---|
Скачать  intel-bug-20121014.f90 | 2.23 КБ |
Thank you for reporting this issue Neil and for the convenient reproducer. I confirmed your results and reported this to the Fortran Developers (internal tracking id noted below) for further analysis/repair. I will update this thread after learning more.
(Internal tracking id: DPD200237446)
(Resolution Update on 02/08/2013): This defect is fixed in the Intel Fortran Composer XE 2013 Update 2 (2013.2.146 - Linux)
I wanted to share our Developer’s comments w/you, Neil. She is working on a fix and I'll update on availability as it becomes known.
She wrote:
First, kudos to the submitter for such a concise reproducer.
Really, what it needed to fail was the following set of circumstances:
1) a TYPE that contains an allocatable CLASS
2) a function that returns that type (or a type that contains that type)
3) that function needs to be called from multiple program units.
Neil - I confirmed this defect is fixed in our internal builds and expect the fix to be in our next XE2013 update 2 (tentatively in late January). The offending intrinsic assignment now also returns T when using our internal build.
Thank you again for reporting this defect and convenient reproducer.
This defect is fixed in the Intel Fortran Composer XE 2013 Update 2 (2013.1.146 - Linux)