@Brecht Van Lommel (brecht), where does the exit-on-missing-textures belong to in your suggestion? It wouldn't be practical to make render farms to abort on all errors reported via clog (or the severity of certain logs is to be revisited).

In D8665#211859, @Sergey Sharybin (sergey) wrote:

@Brecht Van Lommel (brecht), where does the exit-on-missing-textures belong to in your suggestion? It wouldn't be practical to make render farms to abort on all errors reported via clog (or the severity of certain logs is to be revisited).

My suggestion is to only add --debug-exit-on-error in this patch, which will exit on internal/developer errors.

For missing textures, that would be the future --exit-on-error flag. Making such a flag work will indeed require deeper change to our logging and reports (also see my comments in D8691).

Ok, sounds good and easy to tweak the patch to the suggestion then :)

Please add the clog integration now. It should be quite simple, clog already exits on fatal errors.

It seems to be failing on an invalid driver, which should not be an error. Really it's more a user error than internal error, but the simple fix now is to change it to a warning:

diff --git a/source/blender/blenkernel/intern/anim_sys.c b/source/blender/blenkernel/intern/anim_sys.c
index 8fe57f1..69e70cf 100644
--- a/source/blender/blenkernel/intern/anim_sys.c
+++ b/source/blender/blenkernel/intern/anim_sys.c
@@ -2870,7 +2870,7 @@ void BKE_animsys_eval_driver(Depsgraph *depsgraph, ID *id, int driver_index, FCu
 
       /* set error-flag if evaluation failed */
       if (ok == 0) {
-        CLOG_ERROR(&LOG, "invalid driver - %s[%d]", fcu->rna_path, fcu->array_index);
+        CLOG_WARN(&LOG, "invalid driver - %s[%d]", fcu->rna_path, fcu->array_index);
         driver_orig->flag |= DRIVER_FLAG_INVALID;
       }
     }

Also, in this case we should always print the error before Blender exits, even if no logging was enabled with --log. Otherwise it's unclear why Blender exited.

In D8665#212677, @Jacques Lucke (JacquesLucke) wrote:

• print output in case of crashed cycles test

Actually, I was thinking about the error message not being printed by clog. Which could be solved by raising the log level to at least error when enabling --debug-exit-on-error.

The Cycles test crash printing I would leave unchanged.

Honestly, I don't really understand what you mean. Maybe we talk past each other.

In D8665#212720, @Brecht Van Lommel (brecht) wrote:

Actually, I was thinking about the error message not being printed by clog.

Why do we want that the error is not printed when Blender crashes? Don't you think it is useful to have the error printed to be able to debug the issue? I'm not very familiar with clog and the cycles tests, so maybe I'm missing something obvious.

Hi @Jacques Lucke (JacquesLucke), could you say which os you tried this on? When building with make developer most (python) unit tests fail for me on mac and I believe they fail on linux too. (For context, this was discovered when testing D8507 with different make configurations)

I haven't run these tests in debug builds tbh (usually they take way to long). Memory leaks are also detected in release builds. My guess is that there is some debug-mode only code that logs errors..
Running ctest --verbose -R mytestname might give more information.

In D8665#216260, @Jacques Lucke (JacquesLucke) wrote:

I haven't run these tests in debug builds tbh (usually they take way to long). Memory leaks are also detected in release builds. My guess is that there is some debug-mode only code that logs errors..
Running ctest --verbose -R mytestname might give more information.

Actually make developer gives memory leak error in release too. I reported a bug T80605, but I'm still trying to narrow down the exact cause.