Re: [PATCH] {master} Improve and extend tests on `:=' variable assignments.

[Stefano Lattarini]

On Monday 29 November 2010, Ralf Wildenhues wrote:
> * Stefano Lattarini wrote on Mon, Nov 29, 2010 at 09:09:59PM CET:
> > On Monday 29 November 2010, Ralf Wildenhues wrote:
> > > I approve the patch but ask you to keep that coverage in, now you update
> > > the patch with an unrelated new change whose applicability depends on
> > > completely different factors (namely deciding whether some behavior is
> > > desirable or not),
> > >
> > Why should its applicatibility depend on such a decision?  The testcase
> > just serves to expose the current behaviour explicitly, without telling
> > if it's desirable or not.
> 
> IMVHO an entry in the bug tracker is more applicable for issues that may
> be valid or invalid.
>
You're right that an xfailing testcase without an associated PR might
be mostly useless (so yes, my bad in this case too, for proposing a new
xfailing test without opening an associated PR first).  Notice that,
IMHO, the best thing to do is to get both a PR and a testcase, e.g. as
follows:
 - open a new PR in the bug tracker;
 - add xfailing testcase(s), which can now point to an archived PR
   (thus rendering the test much more useful, and its purpose more
   clear);
 - maybe update the PR to refer to the xfailing testcase(s).

> A commit adding a test that is later removed
> because it's invalid is not so easily searched for a couple of years
> later.  Closed PRs on the other hand are straightforward to search.
> 
This is a very good point in favor of the opening of a PR in the bug
tracker, but IMO it's not a reason not to have *also* an xfailing
testcase.

Regards,
   Stefano