Improve EXPECT_XXX and ASSERT_XXX output in case of failure. #2330
Conversation
digit-google
force-pushed
the
better-test-failure-outputs
branch
2 times, most recently
from
b1fb7bb
to
d2c0dd1
Compare
|
Oh, note that the AppVeyor build fails because of the <atlcore.h> include which is fixed by PR #2329 , and has nothing to do with this patch. |
digit-google
force-pushed
the
better-test-failure-outputs
branch
2 times, most recently
from
ac3b944
to
7b1d430
Compare
|
I had to change the patch after realizing that the expressions were evaluated twice in case of failure, which would result in incorrect runtime behavior. For example, consider: Which does happen in the source code, the |
Ensure that binary comparisons (e.g. EXPECT_EQ) display the actual values of the left and right side operands, and not just their source expression, as in: ``` // For failed EXPECT_EQ(expected, actual) Failure on file.cc:10: expected != actual Left: 10 Right: 20 ``` Also ensure that boolean comparisons (e.g. EXPECT_TRUE) display the expected and actual values too as in: ``` // For failed EXPECT_FALSE(vec.size()) Failure on file.cc:10: vec.size() Expected: false Actual: 1 ``` This uses a small amount of C++11 SFINAE, without needing to change any of the test macro invocations. If a value is not supported, then "<UNPRINTABLE>" will be used instead. Note that pointer comparisons with NULL or 0 will now create compiler warnings or errors. So introduce new EXPECT_NULL() and EXPECT_NO_NULL() macros to deal with these. EXPECT_TRUE() and EXPECT_FALSE() still work properly with pointers though, but the error message for EXPECT_NULL() / EXPECT_NOT_NULL() is slightly clearer.
digit-google
force-pushed
the
better-test-failure-outputs
branch
from
7b1d430
to
bad02b1
Compare
|
Mmm, there are too many subtle compilation issues here. I', going to close this PR for now until I can weed out everything, if that is possible. Sorry about the noise. |
|
IMHO #1562 is the way forward. |
|
Thanks, I was not aware that this was still a thing. The latest version of my branch at digit-google@5e6a105 passes all tests now. I'll keep it as a local change in our own Ninja version for now. Do you know if anyone is working on #1562 at the moment? |
|
I would continue working on it when #1583 gets fixed. No one is working on that though. |
Ensure that binary comparisons (e.g. EXPECT_EQ) display the actual values of the left and right side operands, and not just their source expression, as in:
Also ensure that boolean comparisons (e.g. EXPECT_TRUE) display the expected and actual values too as in:
This uses a small amount of C++11 SFINAE, without needing to change any of the test macro invocations. If a value is not supported, then "" will be used instead.