tng unit test shows errors
The tng unit tests shows errors:
23:03:10 8: TNG library: Cannot read header size. D:\jenkins-build\workspace\Gromacs_Gerrit_5_0@2\f9e84682\gromacs\src\external\tng_io\src\lib\tng_io.c: 818
23:03:10 8: TNG library: Cannot read block header at pos -1. D:\jenkins-build\workspace\Gromacs_Gerrit_5_0@2\f9e84682\gromacs\src\external\tng_io\src\lib\tng_io.c: 11553
Without failing. Not sure whether these errors are actually a problem.
- If they are, they need to fixed and also the unit test fixed so that it fails if an error occur
- If it is an problem, they need to be silent to not confuse.
#1 Updated by Magnus Lundborg almost 3 years ago
- Status changed from New to Accepted
The tests should indeed fail at that stage, since somethings is apparently a problem. I guess this is from running bin/tests/tng_testing, right? Could you give the whole output of the tests so that I can see where it fails. I think I know, but it's good to be sure.
The next question is why the test(s) fail(s).
#5 Updated by Roland Schulz almost 3 years ago
Shouldn't the unit test have failed? I understand that it wasn't an error. But I'm wondering that if an error was (wrongly) detected, it seems odd that the unit test passed. I just want to make sure that if a real error happens in the future it isn't missed because the error isn't propagated incorrectly.
#7 Updated by Mark Abraham almost 3 years ago
Magnus Lundborg wrote:
I think that the TNG unit tests in GROMACS currently only check that there are no fatal errors, but Mark might know more about that.
GROMACS doesn't use the TNG unit tests at all - though there's probably a todo about that somewhere. The integration-style tests for various trajectory handling exercise the TNG code, of course, but don't do anything more interesting than wrap trjconv or such, and report on its return code. Some unit style tests on the GROMACS functions that wrap calls to TNG might well be a good idea, but if there's not constructors for stuff that work decently, that should come first!