fortran: Reuse associated_dummy memory if previously allocated [PR108923]

fortran: Reuse associated_dummy memory if previously allocated [PR108923]

[Mikael Morin]

[-- Attachment #1: Type: text/plain, Size: 738 bytes --]

Hello,

Harald found a testcase with memory still leaking despite my previous 
patch for PR108923.
That patch was fixing a leak caused by absence of memory release, the 
attached patch fixes a leak caused by pointer overwrite.

I haven't investigated why sort_actual is called several times( which 
causes the memory leak) nor tried to avoid that.  Theoretically, one 
could assert that the previous associated_dummy value is the same as the 
one to be written (it should be the same at each sort_actual 
invocation), but I have preferred to silently overwrite, and fix just 
the memory problem.

Manually tested on Harald's testcase (predcom-1.f) and ran the full 
fortran testsuite on x86_64-pc-linux-gnu.

OK for master and 12 and 11?

[-- Attachment #2: 0001-fortran-Reuse-associated_dummy-memory-if-previously-.patch --]
[-- Type: text/x-patch, Size: 1762 bytes --]

From 9b88208ec4130712d33d5c7ed74fc17466624a0c Mon Sep 17 00:00:00 2001
From: Mikael Morin <mikael@gcc.gnu.org>
Date: Sat, 25 Feb 2023 16:27:24 +0100
Subject: [PATCH] fortran: Reuse associated_dummy memory if previously
 allocated [PR108923]

This avoids making the associted_dummy field point to a new memory chunk
if it's already pointing somewhere, in which case doing so would leak the
previously allocated chunk.

gcc/fortran/ChangeLog:

	* intrinsic.cc (get_intrinsic_dummy_arg,
	set_intrinsic_dummy_arg): Rename the former to the latter.
	Remove the return value, add a reference to the lhs as argument,
	and do the pointer assignment inside the function.  Don't do
	it if the pointer is already non-NULL.
	(sort_actual): Update caller.
---
 gcc/fortran/intrinsic.cc | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/gcc/fortran/intrinsic.cc b/gcc/fortran/intrinsic.cc
index 17ee999c3b9..e69e541efe0 100644
--- a/gcc/fortran/intrinsic.cc
+++ b/gcc/fortran/intrinsic.cc
@@ -4259,15 +4259,14 @@ remove_nullargs (gfc_actual_arglist **ap)
 }
 
 
-static gfc_dummy_arg *
-get_intrinsic_dummy_arg (gfc_intrinsic_arg *intrinsic)
+static void
+set_intrinsic_dummy_arg (gfc_dummy_arg *&dummy_arg, gfc_intrinsic_arg *intrinsic)
 {
-  gfc_dummy_arg * const dummy_arg = gfc_get_dummy_arg ();
+  if (dummy_arg == NULL)
+    dummy_arg = gfc_get_dummy_arg ();
 
   dummy_arg->intrinsicness = GFC_INTRINSIC_DUMMY_ARG;
   dummy_arg->u.intrinsic = intrinsic;
-
-  return dummy_arg;
 }
 
 
@@ -4430,7 +4429,7 @@ do_sort:
       if (a == NULL)
 	a = gfc_get_actual_arglist ();
 
-      a->associated_dummy = get_intrinsic_dummy_arg (f);
+      set_intrinsic_dummy_arg (a->associated_dummy, f);
 
       if (actual == NULL)
 	*ap = a;
-- 
2.39.1

Re: fortran: Reuse associated_dummy memory if previously allocated [PR108923]

[Harald Anlauf]

Hi Mikael,

Am 25.02.23 um 17:35 schrieb Mikael Morin:
> Hello,
>
> Harald found a testcase with memory still leaking despite my previous
> patch for PR108923.
> That patch was fixing a leak caused by absence of memory release, the
> attached patch fixes a leak caused by pointer overwrite.
>
> I haven't investigated why sort_actual is called several times( which
> causes the memory leak) nor tried to avoid that.  Theoretically, one
> could assert that the previous associated_dummy value is the same as the
> one to be written (it should be the same at each sort_actual
> invocation), but I have preferred to silently overwrite, and fix just
> the memory problem.
>
> Manually tested on Harald's testcase (predcom-1.f) and ran the full
> fortran testsuite on x86_64-pc-linux-gnu.
>
> OK for master and 12 and 11?

LGTM.  OK for master and 12-branch.

It appears that 11-branch is significantly different in the respective
places.  get_intrinsic_dummy_arg does not exist there, so this patch
seems to not apply there.  Or am I missing something?

Thanks for the patch!

Harald

Re: fortran: Reuse associated_dummy memory if previously allocated [PR108923]

[Mikael Morin]

Le 25/02/2023 à 18:20, Harald Anlauf a écrit :
> Hi Mikael,
> 
> Am 25.02.23 um 17:35 schrieb Mikael Morin:
>> Hello,
>>
>> Harald found a testcase with memory still leaking despite my previous
>> patch for PR108923.
>> That patch was fixing a leak caused by absence of memory release, the
>> attached patch fixes a leak caused by pointer overwrite.
>>
>> I haven't investigated why sort_actual is called several times( which
>> causes the memory leak) nor tried to avoid that.  Theoretically, one
>> could assert that the previous associated_dummy value is the same as the
>> one to be written (it should be the same at each sort_actual
>> invocation), but I have preferred to silently overwrite, and fix just
>> the memory problem.
>>
>> Manually tested on Harald's testcase (predcom-1.f) and ran the full
>> fortran testsuite on x86_64-pc-linux-gnu.
>>
>> OK for master and 12 and 11?
> 
> LGTM.  OK for master and 12-branch.
> 
> It appears that 11-branch is significantly different in the respective
> places.  get_intrinsic_dummy_arg does not exist there, so this patch
> seems to not apply there.  Or am I missing something?
> 
No you're right.  I had forgotten that a simplified patch had been used 
to fix pr97896 on the 11 branch.  Thanks.