Comment 11 for bug 661205

Revision history for this message
Martin Pool (mbp) wrote : Re: [Bug 661205] Re: make check still brittle with doctests

On 5 February 2011 02:11, John Arbash Meinel <email address hidden> wrote:
> Hash: SHA1
> On 2/4/2011 1:27 AM, Martin Pool wrote:
>> On 4 February 2011 10:34, Robert Collins <email address hidden> wrote:
>>> Perhaps you could just set pipefile in Makefile ?
>> s//pipefail
>> This would make it bash-specific, and we're probably generally not
>> running bash, and it would need to be done on the same line, since
>> Make runs each line in a separate process.  But otherwise yes.
>> The thing is if we make it exit nonzero on failure, we should not need
>> to have a pipeline at all, as far as I can see.  Just emit subunit to
>> stdout and pqm should be able to use that both for the progress
>> display and to send to the user if it fails.
> As I recall, wasn't the original issue that the subunit-foo that we were
> running at the end of the pipe didn't exit non-zero for failing tests vs
> corrupted stream?

That's right: a file with no errors in it isn't detected as being a
test failure. But it might have no errors because for whatever reason
we didn't end up running the whole test suite. That's why I think
it's inherently more risky to ignore the error code from bzr, and for
bzr to exit 0 even though it knows some tests failed.