cesco-fran

Thanks you @sandcha for your review and feedbacks!.... I think the speed from the old code come at expense of reliability ... some objects were imported and used throughout different...

As your stats confirm the slowness seam to come from the number of calls... that as you mention make the difference when heavy object like tbs have to rebuild more...

Yes, go to "clearer, more independent, and atomic tests" is exactly one the aim of the PR .. to cope with slowness I agree with other strategies should be consider...

While I was reviewing #984 I realize could make sense have some tbs object with different scope ... In my PR I make tbs functionally scoped so that since tbs...

Thanks @srittau for your feedback. The type you define could be a precise type, witch is different from the corresponding lose one, and for that I think could make sense...