When a test prints output during a test, it does so by adding a newline to the beginning of the line, rather than the end.

Do you have a pointer to the code that does this? I don't see any problem with the output as it currently is. Eg excerpt from build/console.out in qemu-arm test run:

...
# starting inlineasm/asmsum.py
array('l', [100, 200, 300, 400]) 1000
array('b', [10, 20, 30, 40, 50, 60, 70, 80]) 360
b'\x01\x02\x03\x04' 10
inlineasm/asmsum.py: OK
# starting qemu-arm/native_test.py
2       
3       
4   
qemu-arm/native_test.py: OK
614 tests ok.  (82 skipped)
status: 0

I've been using MicroPython's copy of tinytest with custom code that is integrated with MicroPython.

Since MicroPython tests use printf with trailing newlines instead of tinytest macros, the problem probably isn't really a problem with anything currently in MicroPython itself. So, this change will probably cause an empty line before qemu-arm/native_test.py: OK in the existing MicroPython test output, but it will fix the regression (caused by the commit I linked above) for general usage of tinytest.

So, this change will probably cause an empty line before qemu-arm/native_test.py: OK in the existing MicroPython test output,

Hmm, that might make it mildly confusing because it will group the result of one test (this OK line) with the output of the next test (the following lines which are not separated by a blank line).

but it will fix the regression (caused by the commit I linked above) for general usage of tinytest.

How did the output look for "normal use" of tinytest before that commit?

Example code:

#include <tinytest.h>
#include <tinytest_macros.h>

static void test1(void *env) {
    tt_fail_msg("didn't work");
}

static struct testcase_t example_tests[] = {
    { "test1", test1, 0,  NULL, NULL },
    END_OF_TESTCASES
};

static struct testgroup_t test_groups[] = {
    { "example/", example_tests },
    END_OF_GROUPS
};

int main(int argc, const char **argv) {
    return tinytest_main(argc, argv, test_groups);
}

Output before f4ed2df:

example/test1: 
  FAIL test.c:5: didn't work
  [test1 FAILED]
1/1 TESTS FAILED. (0 skipped)

Output after f4ed2df:

# starting example/test1

  FAIL test-pbio.c:5: didn't workexample/test1: FAILED
1/1 TESTS FAILED. (0 skipped)

Output with this PR:

# starting example/test1

  FAIL test-pbio.c:227: didn't work
example/test1: FAILED
1/1 TESTS FAILED. (0 skipped)

I think the fact that it always mentions the test name (e.g. example/test1) in the message would prevent any confusion by having an extra blank line. But if it is really a concern, we could further modify tinytest to use newlines at the end instead of the beginning everywhere.

Ok, thanks for the further info.

One option would be to revert the changes made in f4ed2df to keep tinytest as-is (and work as it was originally designed), then modify the MicroPython-tinytest interface code to just print out a bit of extra info, eg:

--- a/tools/tinytest-codegen.py
+++ b/tools/tinytest-codegen.py
@@ -33,8 +33,10 @@ test_function = (
     "void {name}(void* data) {{\n"
     "  static const char pystr[] = {script};\n"
     "  static const char exp[] = {output};\n"
+    '  printf("\\n");\n'
     "  upytest_set_expected_output(exp, sizeof(exp) - 1);\n"
     "  upytest_execute_test(pystr);\n"
+    '  printf("result: ");\n'
     "}}"
 )

I have update the PR with your suggestion. The effects of the changes on console.out can be seen below. The andor.py test was manually rigged to fail so that we could both possible outcomes for a test.

I tested locally with the qemu-arm port and tests pass. Also, this should not break grep "FAIL" ports/qemu-arm/build/console.out in .travis.py.

Current:

# starting basics/array1.py
array('B', [1, 2, 3]) 3
array('I', [1, 2, 3]) 3
1
3
1 -1
2
True
0
array('i')
False
True
False
False
ValueError
basics/array1.py: OK
# starting basics/andor.py
1
(1,)
()
1
basics/andor.py: FAILED

Proposed:

basics/array1.py: 
array('B', [1, 2, 3]) 3
array('I', [1, 2, 3]) 3
1
3
1 -1
2
True
0
array('i')
False
True
False
False
ValueError
result: OK
basics/andor.py: 
1
(1,)
()
1
result: 
  [basics/andor.py FAILED]

Thanks for updating and for the tests. Looks good now. Merged in fd0ba7b