Without the call to g_test_init, the check in test-qofinstance.cpp
fails on: EXPECT_STREQ( error_message, "[qof_commit_edit()] unbalanced
call - resetting (was -2)" ). The error message is set in the local
fatal_handler which is set in the test with
g_test_log_set_fatal_handler.
The same test also does the g_log_set_handler. to the
test_checked_handler method which is in unittest-support.c. It sets
this up to print out the error, but ignore it (if it does NOT match
the way that test_error_struct is setup). But the way it is setup
does not match, because it is setup with level of critical or fatal,
which does not match critical.
Conclusion: The g_test_log_set_fatal_handler only does something when
g_test_init is called. Conversely, g_log_set_handler only does
something when g_test_init is not called. So, the test code should
never use both, and only one or the other depending on if it is a
google test or glib test.
To fix the code for the conversion to google test I am doing:
1. Remove the g_test_init. (per recomendation)
2. Remove the local fatal_handler, and error_message. (It had been
referenced in one other test, but not used.)
3. Remove the check from the (now unused) error_message, and add the
check for number of hits to the test_error_struct.
4. Fix the log level to just be the expected G_LOG_LEVEL_CRITICAL
(removing the or fatal.)
This now passes and does the same checks as before.
When I debug, I like to use a 3 step method:
1. Fix the issue (which is the above)
2. Now that I know this was a problem, find and fix all similar
problems. (Below.)
a) Looking for other loglevel = with combined levels --> I have
found many. At least some of which are part of glibtest,
possibly all, possibly only in libgnucash/engine.
b) Looking for g_log_set_handler I found many, but suspect ones are
in libgnucash/engine.
c) There are multiple places that call both
g_test_log_set_fatal_handler and
g_test_log_set_fatal_handler. Maybe my interpretation above is
wrong, but I tested in the one case above. If I am right this
code is extra in one way or another.
3. Think about what went wrong and keep it for getting this far.
a) g_log_set_handler should assert/fail if it is called after
g_test_init. This is difficult since we don't want to change
glib. The other solution is to replace g_log_set_handler with a
macro that asserts on the check for g_test_init having been
called, and then calls g_log_set_handler.
b) g_test_log_set_fatal_handler should assert if it is called
without calling g_test_init first.
I'm not sure how to proceed with steps 2 and 3.
Both 3a and 3b would only affect test code, so this may be a good
idea. But, this will inject a bunch of errors into the tests. Those
would have to be fixed as part of this commit. Those fixed would
remove code that is closer for google test instead of closer to glib
test.
Maybe my approach should be as follows:
1. convert all existing test_qof (glib) to google test without adding
more coverage.
2. convert all the test-engine tests to google without adding more
coverage.
3. (Possibly) change the engine tests that use do_test mechanism tests
to google test without adding more coverage.
4. The remaining glib tests are outside of engine.
import-export/aqb/test/test-aqb
import-export/test/test-import-pending-matches
register/ledger-core/test/test-split-register
backend/dbi/test/test-backend-dbi backend/sql/test/test-sqlbe
core-utils/test/test-gnc-glib-utils
I'm not sure if converting these, or fixing these for step 3a, 3b
above, or ignoring this is the right step.
5. Adding more coverage to engine.
On Thu, Jan 22, 2026 at 7:46 PM John Ralls <[email protected]> wrote:
>
>
> > On Jan 22, 2026, at 5:44 AM, Stefan Koch <[email protected]>
> wrote:
> >
> > I think the g_test_init is necessary mainly for the
> g_test_log_set_fatal_handler and/or the g_log_set_handler. So, I think I
> will need it even after I replace the glib random numbers.
> >
>
> Stefan,
>
> No, they’re both implemented in glib/gmessages.c with no reference to the
> test system. In fact you might not need to set the fatal handler if you
> don’t call g_test_init: The first thing it does is set the fatal mask to
> CRITICAL, ERROR, an WARNING. Normally only CRITICAL is fatal.
>
> For an example that uses g_log_set_handler without g_test_init see
> https://github.com/Gnucash/gnucash/blob/stable/libgnucash/backend/xml/test/gtest-load-save-files.cpp
>
> Regards,
> John Ralls
>
>
>
_______________________________________________
gnucash-devel mailing list
[email protected]
https://lists.gnucash.org/mailman/listinfo/gnucash-devel
