Ensure make stops on error from go test

[bboreham]

Changed command separator from ; to &&.

In the absence of this (or something like set -e), make will carry on and report the exit status from the next command, which is generally 0 i.e. success.

[rade]

This does have the disadvantage of stopping on the first failing test set, doesn't it? @dpw

[bboreham]

I just checked, and actually it does the opposite - if an earlier test fails but later one succeeds it will return 0 for the whole lot. Back to the drawing board.

[dpw]

This does have the disadvantage of stopping on the first failing test set, doesn't it? @dpw

Looks like it,

It should probably be restructured so that there is an tests empty target that depends on some per-dir test runner targets. It can probably be done with a static pattern.

[bboreham]

We had two much simpler variants before and after 592be90

The multi-line script came in with #505, to generate a consolidated coverage report.

[bboreham]

Changed to accumulate the error in a variable fail

[rade]

Fixed the cosmetic issue. rebased. merged.