Re: [Patch, fortran] PR90218 - [PDT] ICE: tree check: expected array_type, have record_type in gfc_conv_array_initializer, at fortran/trans-array.c:6071

Hi Harald,

Thanks for your comment. I am looking into the reason why I alighted on
that condition. My notes don't recall the reason. I'll clean up/fix, as
necessary before pushing and will post the final version of the patch.

BTW Happy New Year to you!

Paul


On Mon, 5 Jan 2026 at 20:57, Harald Anlauf <[email protected]> wrote:

> Am 05.01.26 um 9:51 PM schrieb Jerry D:
> > On 1/5/26 8:30 AM, Paul Richard Thomas wrote:
> >> Hi All,
> >>
> >> This PR was largely fixed by preceding patches, insofar as it ran and
> >> produced the expected output, apart from the explicit initialization
> >> expressions for the PDT entities. However, the testcase leaked memory
> >> like a sieve and it has taken a while to sort out a satisfactory fix.
> >>
> >> The chunks in trans-decl.cc implement the fix for the explicit
> >> initialization of PDT entities. This part is straightforward.
> >>
> >> The rest of the patch is devoted to fixing the memory leaks triggered
> >> by the new testcase. The problem was associated with pdt_arrays
> >> embedded in PDT structure constructors, which themselves were embedded
> >> in array constructors that were part of a PDT constructor.... if you
> >> see what I mean :-) The original pdt_arrays are copied into a
> >> destination array, which is implicitly allocated
> >> by gfc_duplicate_allocatable. The allocated memory was being lost in
> >> the subsequent copy of the enclosing array constructor. Rescuing and
> >> freeing the memory is accomplished using the finalization block, which
> >> is then executed after the copy of the array constructor. Note that
> >> these frees are not guarded. I don't believe that there is any
> >> circumstance where this will be an issue but, if required, it would be
> >> easily implemented.
> >>
> >> The chunks in trans-stmt.cc pick up memory leaks in allocation and
> >> deallocation. As a side effect, the leak in pdt_3.f03 is fixed.
> >> PR121972 will be updated accordingly, since pdt_39/70/77.f03 still
> >> leak memory.
> >>
> >> Regtests on FC43/x86_64. OK for mainline
> >>
> >> Paul
> >>
> >
> > Looks Good To Me Paul, also regression tested OK here.
> >
> > Jerry
> >
>
> I wanted to says that I am also fine with the patch,
> but have a minor question:
>
> +static tree
> +gfc_init_default_pdt (gfc_symbol *sym, bool dealloc)
>   {
> -  /* Allowed in the case where all the components have initializers and
> -     there are no LEN components.  */
> -  if (sym->ts.type == BT_DERIVED && sym->ts.u.derived->attr.pdt_type)
> +  stmtblock_t block;
> +  tree tmp;
> +  gfc_component *c;
> +
> +  if (sym->value && sym->value->symtree
> +      && sym->value->symtree->n.sym
> +      && !sym->value->symtree->n.sym->attr.artificial)
>
> Don't you want to check that sym->value is really a variable
> before testing symtree?  I.e.,
>
> if (sym->value && sym->value->expr_type == EXPR_VARIABLE
>      && sym->value->symtree
> ...
>
> Thanks,
> Harald
>
>
>
>