Re: Deciding on a default company setup for BDD tests

Hi,

> I think this is a better long term solution as for many scenarios it may
> be impossible to properly remove entries from the database due to the
> Audit Features we have.

Drupal has a tremendous amount of variation between sites, and lots of
configuration that ends up in the database. This certainly colors my
perspective -- and that's why I think it's important to be able to run
BDD tests on a copy of any production database.

I'm not sure that's the same for LedgerSMB -- but it would certainly
help track down issues if people customize their database in ways we
don't expect.

LedgerSMB has a lot of room for between-company variations too, although the number of variations my be lower than with Drupal (e.g. we don't have things as complex as the "Views" module in Drupal). However, I would very much be in favor of trying to distill minimal reproduction recipes when people report errors; mostly because that helps both to verfy that a bug has been fixed -- something that might need to happen multiple times throughout the course of fixing the bug any way.

What we're really talking about here is how to set up test data --
whether we ship a test database already containing data our tests rely
upon, or have those dependencies created when running the tests.

I pretty strongly advocate the latter -- create the configurations/data
we are testing for at the start of a test run, if they don't already
exist. And make it safe to re-run a test on the same database.

This might be a bit of extra effort to achieve: Since we can't remove some data in the database (e.g. transaction deletion is an absolute no-no), it might not always be possible to re-run a test.

I don't mind cleaning up test data if a test fails in development, but
as long as tests are completing, they should be able to be run multiple
times on the same db.

Well, if we clean up behind succesfully run tests, that could also mean we simply delete the test databases in the cluster. Then, we can run the same tests again and again on the given cluster. I'm thinking we will eventually need different databases because we need different company set-ups to test all available features. However, to start, we need a setup with a CoA, accounts and some data, with which we can get an acceptable testing scope in place.

>> > Additionally, John and I were talking about supporting test infrastructure
>> > and we agree that it would be tremendously helpful to be able to see
>> > screenshots of failing scenarios and maybe to be able to see screenshots of
>> > various points in non-failing tests too. Since Travis storage isn't
>> > persistent, we were thinking that we'd need to collect all screenshots as
>> > "build artifacts" and upload them into an AWS S3 account for inspection.
>>
>> Email to ticket system?
>> Or S3...
> Michael makes a really good point here.
> Perhaps the easiest way of capturing the screenshots is not to use S3,

https://docs.travis-ci.com/user/deployment/s3 seems to indicate we can copy to S3 at the end of a build. I think a "bucket" like structure (S3) is going to be much simpler than something meant for version control. (We don't want to version control the images, I'd guess). Ideally, we'd be able to link the images to a build/scenario/step combination and additionally, have an indication of whether the test failed or not.

> but have a github project (eg: ledgersmb-bdd-results) that we can raise
> a ticket against for failing builds with associated screenshots attached.
> At the same time we could use "git annex" to store all screenshots for a
> test in a new git branch (or just simply a tag) in the
> ledgersmb-bdd-results project repository.
>
> Storing "good" results probably should only be done if a specific flag
> is passed in the PR commit message.
> While all screenshots (good and bad) should be stored if a single test
> fails.

However we store them, I suggest we at least store "good" results for
each release. Especially of screenshots. This will allow comparing
version-on-version, as well as give you a place to go back to see "what
did this look like in version x?"

I'm thinking that if the images are small enough, there shouldn't be much of a problem keeping a *loong* tail of history.

S3 storage seems to be built in to many test runners like Travis, I'm
guessing that's the fastest/easiest to get up and running.

Based on https://docs.travis-ci.com/user/deployment/s3 I think we can cause the uploaders built into Travis to automatically set the buckets in S3 to public, so anybody can inspect the results of a failing test (and the previous succesful one).

The Matrix project uses Jenkins as a test runner, and the runs are
public, so you can access artifacts just by visiting their jenkins
instance, no logins necessary. Can Travis do the same?

I have no experience with the uploaders from Travis (i.e. I don't know if there's a link you can click on when you install this), but I would say that at least part of the use-case is supported (the uploading + publishing part).

Bye,

Erik.

http://efficito.com -- Hosted accounting and ERP.

Robust and Flexible. No vendor lock-in.