Hi Herbert,
On Tue, Sep 29, 2020 at 5:44 AM Herbert herbertkoenig@gmx.net wrote:
Hi all,
let me just throw in the cost curve (Barry Boehm I think). A bug gets exponentially more expensive, the later it's found. And the fact that for x bugs found during testing you can be sure to deliver n*x bugs to the customer. n maybe <<1 but still. Don't know the source of this any more.
Thanks, this accords with my experience. Thank you. It also is a strong argument for continuous integration and tests, *BUT* (and it's a big BUT) only if the test signals can be seen.
So lumping lots of tests together so that the one persistently failing test marks the whole suite as having failed isn't helpful. For example, if our SUnit tests were grouped by package then we could see much sooner when a commit of a package broke something. Right now, with everything lumped together the fact that we have some failing tests (an inevitability in complex projects with lots going on, and long term issues that need long term actions to solve) obscures the test results, and means we're not able to use them as we should. If we broke out the test results in sub groups, and didn't report the overall signal, but emphasised the deltas between successive runs of individual groups, we would see when things break.
Right now the CI for VM builds aggregates in exactly this way so when I look in my email I see that the tests have failed (so what's new"). Well, yesterday I had time to look (I don't always) and found that the builds are OK up until a run of a build of a newspeak VM, and the VM build is OK, it is the Newspeak bootstrap that fals, and since Newspeak (I think I'm right in thinking) is no longer maintained, this is not a surprise.
But instead of simply discarding the Newspeak build to get green tests again, better would be to a) not chain the builds so that an early failure prevents potentially successful subsequent builds, i.e. attempt to build everything b) report to the mailing list a message that specifies which builds have failed, so the subject line would be something like "Travis CI: M builds out of N have failed", not the depressing "the build has failed"
I'm bitten by a workflow of quick commits (we are also using Git) as a QA
person in my day job where they also add hasty reviews to the quick commits. So I kind of cringed when I read Christoph's post :-).
Still a complex or tedious workflow is also bad for quality and productivity so that's the point of the post I fully agree with.
I'm too far away from Squeak but would it be a possibility to have small commits in Git (or elsewhere) and consolidated bigger ones for a proper ancestry? At least we need to be very careful when changing the community process.
We have all the infrastructure we need in Monticello (it supports branching and merging, and multiple repositories). We even have something close to what you suggest in production, with release repositories that allow people to stay with e.g. squeak53, and we can push improvements and bug fixes there. But I agree, we want something like a high-frequency trunk and a low-frequency trunk. So we could implement e.g. treatedtrunk, and, say, automate pushing the 3 day, or 7 day old, trunk to treatedtrunk. It would be easy for those wanting a stable trunk process to stay with the treatedtrunk update stream, and to load from trunk the latest greatest that they really want, since the delta between treatedtrunk and trunk would be small, and the delta would be no more than a few days worth of commits.
Best regards,
Herbert
Am 29.09.20 um 01:38 schrieb Thiede, Christoph:
Maybe for git-based projects, a commit-first, fix-later strategy is
what that culture likes and that system can support, but it's not a good fit for Squeak's trunk.
And that's why - sorry for rolling out my old chestnut again - I still do believe that a git-based workflow could make us as a community more efficient.
I don't question that manual testing is great, and I don't question that quality can increase if you give your patches time for maturing, but when I compare the resulting workflow to the "mainstream" workflow that I can use anywhere on GitHub, I repeatedly have the dissatisfying feeling that the inbox/trunk workflow is so slow that it ruins all the efficiency from the Smalltalkish development workflow (which, however, unquestionably outperforms the "mainstream" workflow in a dead, non-live UI without first-class objects for code and tools!).
This might apply most significantly to small changes that would form a PR of two or three commits in a git project because our inbox workflow does not scale so well for changes of such extent. I do not know how many hours I already have spent on fixing the ancestry of my versions, comparing them to their ancestors, or re-uploading them, but it has definitively been too many hours ...
Did someone ever investigate this question seriously by doing a study or so? I would really find the results interesting.
Best,
Christoph
*Von:* Squeak-dev squeak-dev-bounces@lists.squeakfoundation.org squeak-dev-bounces@lists.squeakfoundation.org im Auftrag von Chris Muller asqueaker@gmail.com asqueaker@gmail.com *Gesendet:* Dienstag, 29. September 2020 01:00 Uhr *An:* The general-purpose Squeak developers list *Betreff:* Re: [squeak-dev] tedious programming-in-the-debugger error needs fixing
On Mon, Sep 28, 2020 at 10:07 AM Eliot Miranda eliot.miranda@gmail.com wrote:
Hi Christoph,
On Fri, Sep 25, 2020 at 7:58 AM Thiede, Christoph < Christoph.Thiede@student.hpi.uni-potsdam.de> wrote:
Hi Eliot,
I'm very sorry for this bug, which was so unnecessary because my image has still a gigantic working copy ...! Tools-ct.988 should fix the issue, I tested it in a fresh trunk image. But it would be probably better if you could test it yourself, too. ;-)
No need to apologise. It's an easy mistake, and you fixed it. As long as we're all patient with each other and take responsibility (Andreas said "if you break it, you fix it") we're going to get along fine and be collectively productive.
The following is not addressed to Christoph or his commit, but to Eliots comment, above: Patience should begin within our development methodology. The words above are correct and sweet, and I agree with them, but I feel the need to caution against the implication that "everything's great as long as you fix it *afterward*." Maybe for git-based projects, a *commit-first, fix-later* strategy is what that culture likes and that system can support, but it's not a good fit for Squeak's trunk. I believe Andreas understood this, and he indicated that "breaking the trunk is generally frowned upon."
When it comes to code less than 24 hours old, no matter how simple it seems, chances are about 80% that a subsequent "oops, sorry" commit will need to follow. With "older," (e.g., even only just 48 hours!) *tested* code, that chance drops significantly. Patience. Restraint. Please. :) Let our methodology put time to work *for* us, by living with our changes for a bit (as, it sounds like, Christoph did!) and witness them work in context. Maybe not this time, but *generally, *you'll have a gist enough to know whether it should be loaded and tested separately in a clean trunk first.
Cheers, Chris