Augment Cuspatial Test Utility to Allow User Specified Abs Error #752
Conversation
isVoid
requested review from
trxcllnt,
zhangjianting and
harrism
and removed request for
trxcllnt and
zhangjianting
isVoid
added
feature request
| @@ -69,6 +95,20 @@ MATCHER(float_matcher, std::string(negation ? "are not" : "are") + " approximate | |||
| return false; | |||
| } | |||
|
|
|||
| MATCHER_P(float_near_matcher, | |||
There was a problem hiding this comment.
The float_matcher is also a "near" matcher, so perhaps float_near_matcher is not the right name. The only different is that we pass abs_error. GTEST ignores the abs error if it is zero. So why not just augment the existing matchers with default abs_error of zero to avoid all the additional code?
There was a problem hiding this comment.
GTEST ignores the abs error if it is zero.
Hmm, it's not that simple.. Your assumption is that using FloatNear(abs_error==0) is the same as using FloatEq.
Technically they go through different code paths.
FLOAT_NEAR(abs_error==0) will go to this:
https://github.com/google/googletest/blob/3026483ae575e2de942db5e760cf95e973308dd5/googlemock/include/gmock/gmock-matchers.h#L1687
which will not use the ULP near equal check but completely resort to abs_err check.
FLOAT_EQ matcher will use this:
https://github.com/google/googletest/blob/3026483ae575e2de942db5e760cf95e973308dd5/googlemock/include/gmock/gmock-matchers.h#L1706
which will use the ULP near equal check.
While we can default abs_error to -1, the default value where gtest use internally to differentiate, I believe that's an implementation detail and it's in nature an invalid input to the matcher.
There was a problem hiding this comment.
If just for naming, we can call it float_eq_by_ulp / float_eq_by_abs_error. As for the user facing API expect_vector_equivalent, I think we should keep it as is and use overloads for user to differentiate.
isVoid
added
5 - Ready to Merge
|
@gpucibot merge |
…nge`, adds support to multilinestring distance (#755) Note, this is the first part of `pairwise_linestring_distance` refactoring, part 1 of PR: #753 Depends on #752 Contributes to #706, #703 Closes #745 Authors: - Michael Wang (https://github.com/isVoid) Approvers: - H. Thomson Comer (https://github.com/thomcom) - Mark Harris (https://github.com/harrism) URL: #755
Description
This PR augments Cuspatial Vector Equality Test Utility with an extra parameter to allow user pass in pre-defined absolute error.
In many cases, cuspatial has to compare results with external libraries such as shapely or gdal. Since cuspatial allows user to pick computation precision based on input data (not a bug!), requiring the result to match external libraries within 4ULP is often too stringent, as external libraries often perform computation in different precisions. Adding the flexibility to specifiy abs error in test utilitiy makes it easy for developer to pick desired error range based on specific use cases.
This PR also introduces a test macro that shows lineno of failed tests.
Checklist