lfortran ICE on accessing parent component in a dummy argument #10978

Activity

1. rouson commented on Apr 6, 2026

   

   rouson

   on Apr 6, 2026

   ContributorAuthor

   More actions

   Correction: although the above workaround resolves this issue in the checked configurations below, it fails on the unchecked one.

   • lfortran 0.60.0-421-ge2c448c79 on a macOS ARM system
   • lfortran 0.6.1 on Ubuntu 24.04 in CI on Formal PR 18
   • in Formal PR 18 same PR using LFortran's container

   Is this a regression or is it possible that the container has an lfortran version older than 0.6.1?
2. certik commented on Sep 27, 2026

   

   certik

   on Sep 27, 2026 – with Claude

   Contributor

   More actions

   This is fixed on current main (and in releases since 0.64.0) by commit 71e4f6d "Fix parent-type-name component access crash" (PR #12116, regression test integration_tests/derived_types_149.f90).

   Here is a small example with the pattern from the issue: accessing a component through the parent-type name (self%parent_t%values_) of a class() dummy inside a submodule procedure.

   module m
     implicit none
     type :: parent_t
       double precision, allocatable :: values_(:)
       integer :: order_ = 0
     end type
     type, extends(parent_t) :: child_t
       integer :: extra_ = 0
     end type
     interface
       pure module function child_over_integer(self, n) result(ratio)
         class(child_t), intent(in) :: self
         integer, intent(in) :: n
         type(child_t) :: ratio
       end function
     end interface
   end module
   
   submodule(m) m_s
     implicit none
   contains
     module procedure child_over_integer
       ratio%parent_t = parent_t(self%parent_t%values_/n, order_=self%order_)
       ratio%extra_ = self%extra_
     end procedure
   end submodule
   
   program ex10978
     use m
     implicit none
     type(child_t) :: c, r
     c%parent_t = parent_t([2d0, 4d0, 6d0], order_=2)
     c%extra_ = 7
     r = child_over_integer(c, 2)
     print *, r%values_, r%order_, r%extra_
     if (any(r%values_ /= [1d0, 2d0, 3d0])) error stop
     if (r%order_ /= 2 .or. r%extra_ /= 7) error stop
   end program

   With LFortran 0.63.0 the compiler segfaults:

   $ lfortran ex10978.f90
   Segmentation fault
   

   With current main (0.66.0-445-g611fbfe8d):

   $ lfortran ex10978.f90
   1.0000000000000000 2.0000000000000000 3.0000000000000000 2 7
   

   This matches gfortran.

   I also checked Formal itself with current main, both before the workaround (420c46f) and at the workaround commit (7bb94ce). Both build, and fpm test --compiler lfortran --flag "--cpp --realloc-lhs-arrays --separate-compilation" passes. Current Formal main passes as well.

   About the earlier comment that the workaround still failed in the LFortran container: I could not reproduce that with 0.61, 0.63 or current main. The container image was most likely older than the fix. If you still see a failure with a current ghcr.io/lfortran/lfortran:latest, please post the full lfortran --version, the exact command line, and the error output, and we'll reopen.

   ---

   Generated by Claude Code
3. certik commented on Sep 27, 2026

   

   certik

   on Sep 27, 2026

   Contributor

   More actions

   It looks like this is now fixed.
4. certik

   closed this as completedon Sep 27, 2026