Opened 8 years ago

Last modified 8 years ago

#4779 new Tasks

real test expected failure should be marked as such

Reported by: Jim Bell <jim@…> Owned by: Joel de Guzman
Milestone: Boost 1.45.0 Component: spirit
Version: Boost Development Trunk Severity: Regression
Keywords: Cc:

Description

The qi/real.cpp test failure is described as expected in source-code comments, and indeed fails on every platform.

..\libs\spirit\test\qi\real.cpp(421): test 'test_attr("123233.4124", float_, f) && f == 123233.4140625f' failed in function 'int main()'
1 error detected.

This failure reduces the value of the other tests in qi/real that do pass, as the whole test is marked failed.

Further, this failure causes the entire spirit library to be marked as fail/yellow across every platform.

I suggest splitting the expected failure into a separate test and marking it as an expected failure (see the regression test legend at the bottom of the page). This would better reflect the state of both qi/real and the entire spirit library.

Also, shouldn't the local compare() function be used here instead of operator==()?

Change History (4)

comment:1 Changed 8 years ago by viboes

I don't see this as a showstopper. Please could you set it as a Problem?

comment:2 Changed 8 years ago by Jim Bell <jim@…>

Severity: ShowstopperProblem

Not a showstopper, but I'd like like the next release pass as many tests as possible.

comment:3 Changed 8 years ago by Jim Bell <jim@…>

Severity: ProblemRegression

comment:4 Changed 8 years ago by Hartmut Kaiser

I rolled back the offending test on the release branch for now. This removes the yellow boxes there. We'll attend to that problem before the next release.

Note: See TracTickets for help on using tickets.