Do you mean that restoring the call to PROFILE_END() solve the problem?

I didn't try. But FWICS PROFILE_End() is the only function that frees MRUProfile. I can try if you like.

Yes please. As a matter of fact, I don't know what MRUProfile is and how it is connected to these CONNECT tests. Besides, what make you think that it is the item that is not freed?

Note that because the connect_done_func is only called when the system is shutting down, this memory leak may not be really important. This once more demonstrates the stupidity of these signal raised by some ASSERT MariaDB is full of. ASSERTs, according to all programming experts, is a bad habit that in fact hides the errors, lazily avoid to seach for the cause of them, and prevents to fix them or at least to produce an explicite error message.

I confirm that uncommenting PROFILE_End() in connect_done_func() solves the problem. However I had to also comment out "CurProfile = MRUProfile[i];" and "PROFILE_ReleaseFile();" in PROFILE_End() to make it work properly.

Besides, what make you think that it is the item that is not freed?
"Valgrind warnings" above point to "PROFILE_Open (inihandl.c:455)", which is "MRUProfile[i] = malloc(sizeof(PROFILE));"

Note that because the connect_done_func is only called when the system is shutting down, this memory leak may not be really important.
Strictly speaking connect_done_func() doesn't mean system shutdown, it doesn't even mean MariaDB server shutdown. It means CONNECT engine shutdown. Technically it might be not that important. But it costs us reputation: even if we suppress this leak for valgrind, sooner or later users will find this leak with different memory analysis tool.

This once more demonstrates the stupidity of these signal raised by some ASSERT MariaDB is full of.
How do asserts relate to this bug?

ASSERTs, according to all programming experts, is a bad habit that in fact hides the errors
How is it possible to "hide an error" with assert?

The point of asserts in MariaDB is not "hidding errors" or "avoiding to search for them". Asserts issue additional checks in debug builds, which for performance reason are not issued in release builds.

Good and thank you. I shall do this and hopefully fix this bug.

inihandl.c: As a matter of fact, this code is meant to emulate Windows functions not existing on Unix/Linux. It was imported ten years ago in a previous program of mine and I don't know who wrote it. I have never tried to understand in detail what it does, working almost only on Windows.

Signal was the reason why I commented out this line but apparently this did not happen when you restored the previous version so the reason why it raised a signal has probably been fixed (perhaps by your commenting out these two lines) Perhaps this stuff could be just removed on Linux distributions.

About assertion see for example https://www.softwariness.com/articles/assertions-in-cpp/

Thanks for accepting it. Please also note that MRUProfile[0] can be NULL and other array elements can be uninitialized in PROFILE_End(). The change should look something like this:

diff --git a/storage/connect/inihandl.c b/storage/connect/inihandl.c

index 542b807..9ec566c 100644

--- a/storage/connect/inihandl.c

+++ b/storage/connect/inihandl.c

@@ -622,13 +622,14 @@ void PROFILE_End(void)

   if (trace)

     htrc("PROFILE_End: CurProfile=%p N=%d\n", CurProfile, N_CACHED_PROFILES);

+  if (!CurProfile)

+    return;

+

   /* Close all opened files and free the cache structure */

   for (i = 0; i < N_CACHED_PROFILES; i++) {

     if (trace)

       htrc("MRU=%s i=%d\n", SVP(MRUProfile[i]->filename), i);

-    CurProfile = MRUProfile[i];

-    PROFILE_ReleaseFile();

     free(MRUProfile[i]);

     } // endfor i

Thanks to Sergey Vojtovich