I have dug into this a bit, and found a likely culprit:
The hash table is being created at droute.c: 107
new_path->properties = g_hash_table_new_full ((GHashFunc)str_pair_hash,
str_pair_equal,
g_free,
NULL);
Sure enough, the value deallocator is set to NULL. I am sure there is a
good reason why it is not currently set to g_free, so I don't think that
would be a fix. (although, in this particular memory leak situation,
the "pair" is being allocated with g_new, so using g_free would work).
Perhaps, the values could be wrapped in Glib::RefPtr or std::shared_ptr, or even the simpler std::unique_ptr.
I have CC'd the at-spi group, to get their feedback. On 10/02/2013 10:47 PM, Andrew Potter wrote:
You are using gtkmm correctly. This isn't a gtkmm bug but is in at-spi. I fixed a similar bug for them before (#688363). I'd suggest filing a bug report with a patch on product at-spi if you can. On Wed, Oct 2, 2013 at 10:07 PM, Norman Goldstein <normvcr telus net> wrote:I have searched through the past year, and there are several posts about memory leaks in gtkmm/gtk+. However, I am not clear as to what the recommendation is. I am using gtkmm30 on fedora 18 x86 32-bit. I built and ran a standard piece of sample code: #include <gtkmm.h> int main(int argc, char *argv[]) { Glib::RefPtr<Gtk::Application> app = Gtk::Application::create(argc, argv, "org.gtkmm.examples.base"); Gtk::Window window; window.set_default_size(200, 200); return app->run(window); }// main ***** Is there a clean-up routine I should also be invoking? ******** Valgrind gave the following summary: ==24123== LEAK SUMMARY: ==24123== definitely lost: 280 bytes in 28 blocks ==24123== indirectly lost: 0 bytes in 0 blocks ==24123== possibly lost: 747,922 bytes in 6,535 blocks ==24123== still reachable: 413,852 bytes in 7,213 blocks ==24123== suppressed: 0 bytes in 0 blocks ==24123== Reachable blocks (those to which a pointer was found) are not shown. ==24123== To see them, rerun with: --leak-check=full --show-reachable=yes ==24123== ==24123== For counts of detected and suppressed errors, rerun with: -v ==24123== ERROR SUMMARY: 1085 errors from 1085 contexts (suppressed: 0 from 0) Here is the first "definitely lost" report: ==24123== 8 bytes in 1 blocks are definitely lost in loss record 1,936 of 7,107 ==24123== at 0x4008FAF: malloc (/builddir/build/BUILD/valgrind-3.8.1/coregrind/m_replacemalloc/vg_replace_malloc.c:270) ==24123== by 0x44FF50DB: standard_malloc (/usr/src/debug/glib-2.34.2/glib/gmem.c:85) ==24123== by 0x44FF5481: g_malloc (/usr/src/debug/glib-2.34.2/glib/gmem.c:159) ==24123== by 0x476A8C06: droute_path_add_interface (/usr/src/debug/at-spi2-atk-2.6.2/droute/droute.c:234) (<=================) ==24123== by 0x476B0845: spi_initialize_selection (/usr/src/debug/at-spi2-atk-2.6.2/atk-adaptor/adaptors/selection-adaptor.c:262) ==24123== by 0x476A333D: atk_bridge_adaptor_init (/usr/src/debug/at-spi2-atk-2.6.2/atk-adaptor/bridge.c:907) ==24123== by 0x41D4C3C2: _gtk_accessibility_init (/usr/src/debug/gtk+-3.6.4/gtk/a11y/gail.c:831) ==24123== by 0x41BAA546: post_parse_hook (/usr/src/debug/gtk+-3.6.4/gtk/gtkmain.c:733) ==24123== by 0x44FFB429: g_option_context_parse (/usr/src/debug/glib-2.34.2/glib/goption.c:2001) ==24123== by 0x41BAAAE7: gtk_parse_args (/usr/src/debug/gtk+-3.6.4/gtk/gtkmain.c:952) ==24123== by 0x41BAAB73: gtk_init_check (/usr/src/debug/gtk+-3.6.4/gtk/gtkmain.c:991) ==24123== by 0x41BAABB3: gtk_init (/usr/src/debug/gtk+-3.6.4/gtk/gtkmain.c:1046) I installed the debuginfo packages to get the source file information. The leak is occurring at droute.c (indicated by the arrow, above): pair = g_new (PropertyPair, 1); //<===== Line 234 pair->get = properties->get; pair->set = properties->set; g_hash_table_insert (path->properties, str_pair_new (itf, prop), pair); //<====== The first arrow is where "pair" is allocated. It gets stored into a hash table at the 2nd arrow. Either there is a duplicate insertion that is not being managed (deleting the original insertion), or the table is not being destroyed, or it is being destroyed but not managing its data. Most likely, I don't know the clean way of shutting down a gtkmm/gtk+ session. _______________________________________________ gtkmm-list mailing list gtkmm-list gnome org https://mail.gnome.org/mailman/listinfo/gtkmm-list
Attachment:
smime.p7s
Description: S/MIME Cryptographic Signature