On Mon, May 24, 2021 at 10:09:38PM +0200, Christian Weisgerber wrote:
> Stefan Sperling:
> 
> > As a first step, I'd like to commit the test case below as an 'xfail' test,
> > even though the embedded test data is C code which looks ugly and confusing
> > when embedded into a shell script like this...
> 
> Well, I don't see why a test couldn't have an extra data file.
> Or you could start each line with some marker character and trim
> it off before use:
>   sed -e 's/^X//' > $testroot/repo/reference.c <<EOF
> 
> But I think the current embedding is fine.

Fair enough, I'll leave it as it is.
Hopefully a better test case will be found eventually.

> > +	if [ "$ret" != "0" ]; then
> > +		diff -u $testroot/diff.expected $testroot/diff
> > +		ret="xfail cherrypick results in unexpected diff"
> > +		test_done "$testroot" "$ret"
> > +		return 1
> > +	fi
> 
> Shouldn't an xfail NOT return failure, i.e., return 0?

Indeed. We don't want 'make regress' to flag this as a failure.
Thank you, I will fix this.