Re: pushing patches despite requests for a test

Hey Michael,

I hope you don't mind me butting in again.

On Sat, 2015-07-25 at 08:17 +0100, Michael Meeks wrote:
> Hi Markus,
>
> On Fri, 2015-07-24 at 18:22 +0200, Markus Mohrhard wrote:
> > so it is now the second time that despite me requesting a unit test in
> > a gerrit review request a patch has been pushed.
>
>       Sounds like bad style. Then again - how many man-hours do we expect
> would be required for the tests ? [ if it is easy to test then ... worse
> style I guess ].
>
> > I only care marginally if you do it in code that I don't maintain but
> > I will revert it every single time when it is in code that I maintain.
>
>       I wonder what the wider context is; I imagine people are fixing crazily
> for -5-0-0 - and that in some cases creating a unit test consumes
> significant time that will stop the next fix being got at

This is the wrong way to look at it in my opinion.  It's a short-term
small gain, and long-term disaster, not to mention this does not help
promote the right mindset amongst the developers that tests are far more
important than the fixes themselves (again in my opinion).  It's a
long-term disaster because more often than not, people forget to write
the test that one promised to write (or stop caring to write one after
some cool down period).

>  - which will
> ultimately result in a noticeably poorer quality 5.0.0 release. ie.
> we're in a short-term bug-fix crunch and this is a zero sum game to some
> extent.

The goal I cared more about when I was still active was a long-term
stability than making the next release "stable", which in my mind
aligned very well with the time-based release philosophy that the
project decided to adopt at its inception i.e. if the bug fix misses the
next release such as 5.0.0, there is always 5.0.1 to look forward to.

But then again, I soon realized that only a few other people shared my
view, which was bit disappointing.

>       Then again, it sounds unhelpful longer term; I wonder if we could have
> the fixes on the -5-0-0 branch but not on master or on -5-0 (without a
> unit test) - which would of course be pretty 'orrible as an approach:
> but hopefully queue up the unit testing work to make sure that it gets
> done later & yet get the fix in now.

So, the way I look at this is that there is no "easy way out".  My
impression is that almost everyone (except Markus) is trying to find a
easier and clever way out.  But in my view the test-first philosophy is not an easy
life, but something we all have to strive for in order to maintain
quality while at the same time encouraging the developers to keep making
interesting and exciting changes to the code base.  It's a hard life,
but it needs to be done.

Again, all of this is strictly my opinion.  So, feel free to "just take
it or leave it".

Kohei