Skip to contentSkip to conversation

tests/qtest: Plug memory leaks in qtest_get_machines

PATCH In review Fabiano Rosas 2 participants 8 messages Jan 20, 2023 – Jan 24, 2023

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 tests/qtest: Plug memory leaks in qtest_get_machines

These leaks can be avoided:

759 bytes in 61 blocks are still reachable in loss record 56 of 60

    at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
    by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
    by 0x4AA313E: g_strdup (in /usr/lib64/libglib-2.0.so.0.7000.5)
    by 0x12083E: qtest_get_machines (libqtest.c:1323)
    by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
    by 0x11556C: main (test-hmp.c:160)

992 bytes in 1 blocks are still reachable in loss record 57 of 60

    at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
    by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
    by 0x120725: qtest_get_machines (libqtest.c:1313)
    by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
    by 0x11556C: main (test-hmp.c:160)
---
 tests/qtest/libqtest.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
tests/qtest/libqtest.c +14 -0
index 6b2216cb20..65abac5029 100644
@@ -1285,6 +1285,18 @@ struct MachInfo {
     char *alias;
 };
 
+static void qtest_free_machine_info(gpointer data)
+{
+    struct MachInfo *machines = data;
+    int i;
+
+    for (i = 0; machines[i].name != NULL; i++) {
+        g_free((void *)machines[i].name);
+        g_free((void *)machines[i].alias);
+    }
+    g_free(machines);
+}
+
 /*
  * Returns an array with pointers to the available machine names.
  * The terminating entry has the name set to NULL.
@@ -1336,6 +1348,8 @@ static struct MachInfo *qtest_get_machines(void)
     qobject_unref(response);
 
     memset(&machines[idx], 0, sizeof(struct MachInfo)); /* Terminating entry */
+    g_test_queue_destroy(qtest_free_machine_info, machines);
+
     return machines;
 }
 
Signature · 1 line
2.35.3
#1 lore raw
Quoted 27 lines · Fabiano Rosas

These leaks can be avoided:

759 bytes in 61 blocks are still reachable in loss record 56 of 60

     at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
     by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
     by 0x4AA313E: g_strdup (in /usr/lib64/libglib-2.0.so.0.7000.5)
     by 0x12083E: qtest_get_machines (libqtest.c:1323)
     by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
     by 0x11556C: main (test-hmp.c:160)

992 bytes in 1 blocks are still reachable in loss record 57 of 60

     at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
     by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
     by 0x120725: qtest_get_machines (libqtest.c:1313)
     by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
     by 0x11556C: main (test-hmp.c:160)
---
  tests/qtest/libqtest.c | 14 ++++++++++++++
  1 file changed, 14 insertions(+)
tests/qtest/libqtest.c +0 -0
index 6b2216cb20..65abac5029 100644

tests/qtest/libqtest.c struct MachInfo { · line 1294

Quoted 10 lines
tests/qtest/libqtest.c +7 -0
@@ -1285,6 +1285,18 @@ struct MachInfo {
      char *alias;
  };
 
+static void qtest_free_machine_info(gpointer data)
+{
+    struct MachInfo *machines = data;
+    int i;
+
+    for (i = 0; machines[i].name != NULL; i++) {
+        g_free((void *)machines[i].name); > +        g_free((void *)machines[i].alias);

I'd suggest setting .name and .alias to NULL after freeing them, to avoid that danling pointers are left behind.

tests/qtest/libqtest.c static struct MachInfo *qtest_get_machines(void) · line 1351

Quoted 12 lines
tests/qtest/libqtest.c +5 -0
+    }
+    g_free(machines);
+}
+
  /*
   * Returns an array with pointers to the available machine names.
   * The terminating entry has the name set to NULL.
@@ -1336,6 +1348,8 @@ static struct MachInfo *qtest_get_machines(void)
      qobject_unref(response);
 
      memset(&machines[idx], 0, sizeof(struct MachInfo)); /* Terminating entry */
+    g_test_queue_destroy(qtest_free_machine_info, machines);

So this frees the machines structure...

tests/qtest/libqtest.c static struct MachInfo *qtest_get_machines(void) · line 1353

Quoted 1 line
tests/qtest/libqtest.c +0 -0
      return machines;

... but here it gets returned, too? ... that looks wrong. Did you maybe rather want to free it at the end of qtest_cb_for_every_machine() and qtest_has_machine ?

Thomas

Quoted 2 lines
tests/qtest/libqtest.c +0 -0
  }
 
#2 lore raw
Quoted 57 lines · Thomas Huth
Quoted 27 lines · Fabiano Rosas

These leaks can be avoided:

759 bytes in 61 blocks are still reachable in loss record 56 of 60

     at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
     by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
     by 0x4AA313E: g_strdup (in /usr/lib64/libglib-2.0.so.0.7000.5)
     by 0x12083E: qtest_get_machines (libqtest.c:1323)
     by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
     by 0x11556C: main (test-hmp.c:160)

992 bytes in 1 blocks are still reachable in loss record 57 of 60

     at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
     by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
     by 0x120725: qtest_get_machines (libqtest.c:1313)
     by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
     by 0x11556C: main (test-hmp.c:160)
---
  tests/qtest/libqtest.c | 14 ++++++++++++++
  1 file changed, 14 insertions(+)
tests/qtest/libqtest.c +0 -0
index 6b2216cb20..65abac5029 100644

tests/qtest/libqtest.c struct MachInfo { · line 1294

Quoted 10 lines
tests/qtest/libqtest.c +7 -0
@@ -1285,6 +1285,18 @@ struct MachInfo {
      char *alias;
  };
 
+static void qtest_free_machine_info(gpointer data)
+{
+    struct MachInfo *machines = data;
+    int i;
+
+    for (i = 0; machines[i].name != NULL; i++) {
+        g_free((void *)machines[i].name); > +        g_free((void *)machines[i].alias);

I'd suggest setting .name and .alias to NULL after freeing them, to avoid that danling pointers are left behind.

tests/qtest/libqtest.c static struct MachInfo *qtest_get_machines(void) · line 1351

Quoted 12 lines
tests/qtest/libqtest.c +5 -0
+    }
+    g_free(machines);
+}
+
  /*
   * Returns an array with pointers to the available machine names.
   * The terminating entry has the name set to NULL.
@@ -1336,6 +1348,8 @@ static struct MachInfo *qtest_get_machines(void)
      qobject_unref(response);
 
      memset(&machines[idx], 0, sizeof(struct MachInfo)); /* Terminating entry */
+    g_test_queue_destroy(qtest_free_machine_info, machines);

So this frees the machines structure...

Quoted 5 lines

tests/qtest/libqtest.c static struct MachInfo *qtest_get_machines(void) · line 1353

Quoted 1 line
tests/qtest/libqtest.c +0 -0
      return machines;

... but here it gets returned, too? ... that looks wrong. Did you maybe rather want to free it at the end of qtest_cb_for_every_machine() and qtest_has_machine ?

g_test_queue_destroy will only call qtest_free_machine_info during the test teardown phase:

#0  qtest_free_machine_info (data=0x555555677870) at ../tests/qtest/libqtest.c:1289
#1  0x00007ffff7b1d9d1 in ?? () from /usr/lib64/libglib-2.0.so.0
#2  0x00007ffff7b1d8b3 in ?? () from /usr/lib64/libglib-2.0.so.0
#3  0x00007ffff7b1d8b3 in ?? () from /usr/lib64/libglib-2.0.so.0
#4  0x00007ffff7b1de82 in g_test_run_suite () from /usr/lib64/libglib-2.0.so.0
#5  0x00007ffff7b1deab in g_test_run () from /usr/lib64/libglib-2.0.so.0
#6  0x0000555555561221 in main (argc=<optimized out>, argv=<optimized

#out>) at ../tests/qtest/qom-test.c:12

As long as 'machines' is static and not being exposed to the tests, I think this should be fine.

#3 lore raw
Quoted 75 lines · Fabiano Rosas
Quoted 57 lines · Thomas Huth
Quoted 27 lines · Fabiano Rosas

These leaks can be avoided:

   759 bytes in 61 blocks are still reachable in loss record 56 of 60
      at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
      by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
      by 0x4AA313E: g_strdup (in /usr/lib64/libglib-2.0.so.0.7000.5)
      by 0x12083E: qtest_get_machines (libqtest.c:1323)
      by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
      by 0x11556C: main (test-hmp.c:160)

   992 bytes in 1 blocks are still reachable in loss record 57 of 60
      at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
      by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
      by 0x120725: qtest_get_machines (libqtest.c:1313)
      by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
      by 0x11556C: main (test-hmp.c:160)
---
   tests/qtest/libqtest.c | 14 ++++++++++++++
   1 file changed, 14 insertions(+)
tests/qtest/libqtest.c +0 -0
index 6b2216cb20..65abac5029 100644

tests/qtest/libqtest.c struct MachInfo { · line 1294

Quoted 10 lines
tests/qtest/libqtest.c +7 -0
@@ -1285,6 +1285,18 @@ struct MachInfo {
       char *alias;
   };
 
+static void qtest_free_machine_info(gpointer data)
+{
+    struct MachInfo *machines = data;
+    int i;
+
+    for (i = 0; machines[i].name != NULL; i++) {
+        g_free((void *)machines[i].name); > +        g_free((void *)machines[i].alias);

I'd suggest setting .name and .alias to NULL after freeing them, to avoid that danling pointers are left behind.

tests/qtest/libqtest.c static struct MachInfo *qtest_get_machines(void) · line 1351

Quoted 12 lines
tests/qtest/libqtest.c +5 -0
+    }
+    g_free(machines);
+}
+
   /*
    * Returns an array with pointers to the available machine names.
    * The terminating entry has the name set to NULL.
@@ -1336,6 +1348,8 @@ static struct MachInfo *qtest_get_machines(void)
       qobject_unref(response);
 
       memset(&machines[idx], 0, sizeof(struct MachInfo)); /* Terminating entry */
+    g_test_queue_destroy(qtest_free_machine_info, machines);

So this frees the machines structure...

Quoted 5 lines

tests/qtest/libqtest.c static struct MachInfo *qtest_get_machines(void) · line 1353

Quoted 1 line
tests/qtest/libqtest.c +0 -0
       return machines;

... but here it gets returned, too? ... that looks wrong. Did you maybe rather want to free it at the end of qtest_cb_for_every_machine() and qtest_has_machine ?

g_test_queue_destroy will only call qtest_free_machine_info during the test teardown phase:

#0  qtest_free_machine_info (data=0x555555677870) at ../tests/qtest/libqtest.c:1289
#1  0x00007ffff7b1d9d1 in ?? () from /usr/lib64/libglib-2.0.so.0
#2  0x00007ffff7b1d8b3 in ?? () from /usr/lib64/libglib-2.0.so.0
#3  0x00007ffff7b1d8b3 in ?? () from /usr/lib64/libglib-2.0.so.0
#4  0x00007ffff7b1de82 in g_test_run_suite () from /usr/lib64/libglib-2.0.so.0
#5  0x00007ffff7b1deab in g_test_run () from /usr/lib64/libglib-2.0.so.0
#6  0x0000555555561221 in main (argc=<optimized out>, argv=<optimized
Quoted 4 lines

#out>) at ../tests/qtest/qom-test.c:12

As long as 'machines' is static and not being exposed to the tests, I think this should be fine.

Ah, thanks for the explanation, I really got that wrong.

But I think g_test_queue_destroy() might still be the wrong thing to use here. I added a fprintf() to your new qtest_free_machine_info() funcion, and it seems to be called once at the end of the very first test already. So if anything else calls qtest_get_machines() again after the first test finished, it will crash due to the dangling static machine pointer.

So maybe the static machine pointer should be moved outside of the function and then be released from qtest_quit() instead?

(Also, it's valgrind that you used, isn't it? - I wonder why it's complaining here at all since the pointer to the memory should still be valid at the end?)

Thomas

#4 lore raw
Quoted 78 lines · Thomas Huth
Quoted 77 lines · Fabiano Rosas
Quoted 57 lines · Thomas Huth
Quoted 27 lines · Fabiano Rosas

These leaks can be avoided:

   759 bytes in 61 blocks are still reachable in loss record 56 of 60
      at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
      by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
      by 0x4AA313E: g_strdup (in /usr/lib64/libglib-2.0.so.0.7000.5)
      by 0x12083E: qtest_get_machines (libqtest.c:1323)
      by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
      by 0x11556C: main (test-hmp.c:160)

   992 bytes in 1 blocks are still reachable in loss record 57 of 60
      at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
      by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
      by 0x120725: qtest_get_machines (libqtest.c:1313)
      by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
      by 0x11556C: main (test-hmp.c:160)
---
   tests/qtest/libqtest.c | 14 ++++++++++++++
   1 file changed, 14 insertions(+)
tests/qtest/libqtest.c +0 -0
index 6b2216cb20..65abac5029 100644

tests/qtest/libqtest.c struct MachInfo { · line 1294

Quoted 10 lines
tests/qtest/libqtest.c +7 -0
@@ -1285,6 +1285,18 @@ struct MachInfo {
       char *alias;
   };
 
+static void qtest_free_machine_info(gpointer data)
+{
+    struct MachInfo *machines = data;
+    int i;
+
+    for (i = 0; machines[i].name != NULL; i++) {
+        g_free((void *)machines[i].name); > +        g_free((void *)machines[i].alias);

I'd suggest setting .name and .alias to NULL after freeing them, to avoid that danling pointers are left behind.

tests/qtest/libqtest.c static struct MachInfo *qtest_get_machines(void) · line 1351

Quoted 12 lines
tests/qtest/libqtest.c +5 -0
+    }
+    g_free(machines);
+}
+
   /*
    * Returns an array with pointers to the available machine names.
    * The terminating entry has the name set to NULL.
@@ -1336,6 +1348,8 @@ static struct MachInfo *qtest_get_machines(void)
       qobject_unref(response);
 
       memset(&machines[idx], 0, sizeof(struct MachInfo)); /* Terminating entry */
+    g_test_queue_destroy(qtest_free_machine_info, machines);

So this frees the machines structure...

Quoted 5 lines

tests/qtest/libqtest.c static struct MachInfo *qtest_get_machines(void) · line 1353

Quoted 1 line
tests/qtest/libqtest.c +0 -0
       return machines;

... but here it gets returned, too? ... that looks wrong. Did you maybe rather want to free it at the end of qtest_cb_for_every_machine() and qtest_has_machine ?

g_test_queue_destroy will only call qtest_free_machine_info during the test teardown phase:

#0  qtest_free_machine_info (data=0x555555677870) at ../tests/qtest/libqtest.c:1289
#1  0x00007ffff7b1d9d1 in ?? () from /usr/lib64/libglib-2.0.so.0
#2  0x00007ffff7b1d8b3 in ?? () from /usr/lib64/libglib-2.0.so.0
#3  0x00007ffff7b1d8b3 in ?? () from /usr/lib64/libglib-2.0.so.0
#4  0x00007ffff7b1de82 in g_test_run_suite () from /usr/lib64/libglib-2.0.so.0
#5  0x00007ffff7b1deab in g_test_run () from /usr/lib64/libglib-2.0.so.0
#6  0x0000555555561221 in main (argc=<optimized out>, argv=<optimized

#out>) at ../tests/qtest/qom-test.c:12

Quoted 8 lines
Quoted 2 lines

As long as 'machines' is static and not being exposed to the tests, I think this should be fine.

Ah, thanks for the explanation, I really got that wrong.

But I think g_test_queue_destroy() might still be the wrong thing to use here. I added a fprintf() to your new qtest_free_machine_info() funcion, and it seems to be called once at the end of the very first test already.

Yes, because we currently only use qtest_get_machines to construct the test paths. So it is only needed before the tests.

The g_test_queue_destroy function is actually attaching the callback to the first test.

Quoted 3 lines

So if anything else calls qtest_get_machines() again after the first test finished, it will crash due to the dangling static machine pointer.

Ah, my reply about it being fine was while looking at the v2 which sets 'machines' to NULL. So we would go around qtest_get_machines once more and register qtest_free_machine_info a second time. Every time we have to fetch the machines, we register the cleanup callback.

Quoted 2 lines

So maybe the static machine pointer should be moved outside of the function and then be released from qtest_quit() instead?

Hm.. let me give it a try. I'm not sure if it works because we call qtest_quit while still using the machines array. Maybe if I move it out and do the cleanup in qtest_cb_for_every_machine...

Quoted 3 lines

(Also, it's valgrind that you used, isn't it? - I wonder why it's complaining here at all since the pointer to the memory should still be valid at the end?)

valgrind is complaining about the memory not being explicitly freed at any point. I guess "leak" was not the most precise term to use in the commit message.

#5 lore raw
Quoted 22 lines · Fabiano Rosas
Quoted 20 lines · Thomas Huth
Quoted 19 lines · Fabiano Rosas
Quoted 17 lines · Thomas Huth
Quoted 16 lines · Fabiano Rosas

These leaks can be avoided:

    759 bytes in 61 blocks are still reachable in loss record 56 of 60
       at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
       by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
       by 0x4AA313E: g_strdup (in /usr/lib64/libglib-2.0.so.0.7000.5)
       by 0x12083E: qtest_get_machines (libqtest.c:1323)
       by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
       by 0x11556C: main (test-hmp.c:160)

    992 bytes in 1 blocks are still reachable in loss record 57 of 60
       at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
       by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
       by 0x120725: qtest_get_machines (libqtest.c:1313)
       by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
       by 0x11556C: main (test-hmp.c:160)

...

Quoted 7 lines
Quoted 3 lines

(Also, it's valgrind that you used, isn't it? - I wonder why it's complaining here at all since the pointer to the memory should still be valid at the end?)

valgrind is complaining about the memory not being explicitly freed at any point. I guess "leak" was not the most precise term to use in the commit message.

How did you run valgrind? For me, it does not really complain about the non-freed memory here since it is still reachable. The only difference that I see is in the summary. Without your patch:

still reachable: 27,471 bytes in 152 blocks

with your patch:

still reachable: 25,713 bytes in 88 blocks

... but that IMHO doesn't really hurt, since the memory is not really leaked, i.e. the memory usage won't increase during runtime here. So I fail to see which problem you're really trying to solve here, could you please elaborate?

Thomas

#6 lore raw
Quoted 36 lines · Thomas Huth
Quoted 22 lines · Fabiano Rosas
Quoted 20 lines · Thomas Huth
Quoted 19 lines · Fabiano Rosas
Quoted 17 lines · Thomas Huth
Quoted 16 lines · Fabiano Rosas

These leaks can be avoided:

    759 bytes in 61 blocks are still reachable in loss record 56 of 60
       at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
       by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
       by 0x4AA313E: g_strdup (in /usr/lib64/libglib-2.0.so.0.7000.5)
       by 0x12083E: qtest_get_machines (libqtest.c:1323)
       by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
       by 0x11556C: main (test-hmp.c:160)

    992 bytes in 1 blocks are still reachable in loss record 57 of 60
       at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
       by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
       by 0x120725: qtest_get_machines (libqtest.c:1313)
       by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
       by 0x11556C: main (test-hmp.c:160)

...

Quoted 7 lines
Quoted 3 lines

(Also, it's valgrind that you used, isn't it? - I wonder why it's complaining here at all since the pointer to the memory should still be valid at the end?)

valgrind is complaining about the memory not being explicitly freed at any point. I guess "leak" was not the most precise term to use in the commit message.

How did you run valgrind? For me, it does not really complain about the non-freed memory here since it is still reachable. The only difference that I see is in the summary. Without your patch:

Quoted 5 lines

still reachable: 27,471 bytes in 152 blocks

with your patch:

still reachable: 25,713 bytes in 88 blocks

valgrind --leak-check=full --show-leak-kinds=all

Quoted 4 lines

... but that IMHO doesn't really hurt, since the memory is not really leaked, i.e. the memory usage won't increase during runtime here. So I fail to see which problem you're really trying to solve here, could you please elaborate?

You're right, its harmless. We could just not bother with it.

#7 lore raw
Quoted 38 lines · Fabiano Rosas
Quoted 36 lines · Thomas Huth
Quoted 22 lines · Fabiano Rosas
Quoted 20 lines · Thomas Huth
Quoted 19 lines · Fabiano Rosas
Quoted 17 lines · Thomas Huth
Quoted 16 lines · Fabiano Rosas

These leaks can be avoided:

     759 bytes in 61 blocks are still reachable in loss record 56 of 60
        at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
        by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
        by 0x4AA313E: g_strdup (in /usr/lib64/libglib-2.0.so.0.7000.5)
        by 0x12083E: qtest_get_machines (libqtest.c:1323)
        by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
        by 0x11556C: main (test-hmp.c:160)

     992 bytes in 1 blocks are still reachable in loss record 57 of 60
        at 0x4034744: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
        by 0x4A88518: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.5)
        by 0x120725: qtest_get_machines (libqtest.c:1313)
        by 0x12098C: qtest_cb_for_every_machine (libqtest.c:1348)
        by 0x11556C: main (test-hmp.c:160)

...

Quoted 7 lines
Quoted 3 lines

(Also, it's valgrind that you used, isn't it? - I wonder why it's complaining here at all since the pointer to the memory should still be valid at the end?)

valgrind is complaining about the memory not being explicitly freed at any point. I guess "leak" was not the most precise term to use in the commit message.

How did you run valgrind? For me, it does not really complain about the non-freed memory here since it is still reachable. The only difference that I see is in the summary. Without your patch:

Quoted 7 lines
Quoted 5 lines

still reachable: 27,471 bytes in 152 blocks

with your patch:

still reachable: 25,713 bytes in 88 blocks

valgrind --leak-check=full --show-leak-kinds=all

Ok, so that "--show-leak-kinds=all" is the issue here. I think it does not make sense to go hunting for each and every "reachable" non-freed memory block - since they have not really been leaked during runtime. It's maybe better if you just use "--show-leak-kinds=definite,indirect,possible" instead.

Thomas