On 11/12/26 4:15 AM, Paul Richard Thomas wrote:
Hello All,
The attached patch fixes the last of the runtime bugs in pr82173 comment #10.
The patch fixes double frees of parameterized components of source
expressions in calls to array transformational functions. Before this
fix, the contents of the parameterized components were shallow copied
to the result. In all cases, a deep copy was required and is
implemented here.
Merge and transfer are inlined and scalarized, which then need the new
copy function to be called from three locations.
The testcase, pdt_bug26.f90 is free of double frees, memory leaks and
errors, according to valgrind.
The patch passes regression testing on FC44/x86_64. OK for mainline?
Patch 15/16, provided by Jerry, will provide a fix for a problem
with--with-arch=armv8-a reported by Linaro.
Patch 16/16 will provide a fix for the 11 missing F2023 constraints for PDTs.
Then I will analyze the existing pdt tests to see if any cover cases
not included in the full comment #10 testsuite. These will be
retained, the rest eliminated and replaced by a consolidated version
of the new testsuite.
After this, apart from regressions, I am done with PDTs forever! One
gratifying feature of this PDT campaign is that the original design,
especially with LEN components, has stood the test of time. Complexity
of implementation has been the problem all along.
Regards
This is OK. I have been having some email problems on my end so if I am
repeating myself, please forgive.
Jerry> Paul
