Re: [PATCH 1/4] line-log: demonstrate a bug with nearly-overlapping ranges

[Johannes Schindelin <Johannes.Schindelin@gmx.de>]

Hi Jonathan,

On Sat, 4 Aug 2018, Jonathan Nieder wrote:

> Johannes Schindelin wrote:
> 
> > Currently, this test case throws an assertion:
> >
> > 	Assertion failed!
> >
> > 	Program: git.exe
> > 	File: line-log.c, Line 71
> >
> > Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
> > ---
> >  t/t4211-line-log.sh | 17 +++++++++++++++++
> >  1 file changed, 17 insertions(+)
> 
> Thanks for finding and demonstrating it.

You're welcome.

> Can you say more about what is going on in the test case?

Certainly. I considered writing more into the commit message, but all my
attempts were really repeating what the test case does, and were therefore
poor commit message material.

Under certain circumstances, multiple ranges specified for the line-log
were adjusted incorrectly, leading to violation of assumptions as well as
to segmentation faults. This test case demonstrates this.

> Alternatively, could it be squashed in with the patch that fixes it?

There is this really awful trend on this mailing list to suggest
conflating the demonstration of a bug with the bug fix.

It is really, really important to realize how valuable it is to have the
regression test as an individual patch that can be used to verify that
there is a bug, to pinpoint where it was introduced, to test alternative
fixes, to keep records separate, and I could go on and on and on. Please
do not ignore these very good reasons, and please refrain from
recommending such conflation in the future.

> The below will be more nitpicky:
> 
> [...]
> > --- a/t/t4211-line-log.sh
> > +++ b/t/t4211-line-log.sh
> > @@ -115,4 +115,21 @@ test_expect_success 'range_set_union' '
> >  	git log $(for x in $(test_seq 200); do echo -L $((2*x)),+1:c.c; done)
> >  '
> >  
> > +q_to_lf () {
> > +	tr Q '\012'
> > +}
> > +
> > +test_expect_failure 'close to overlapping ranges' '
> > +	test_seq 5 >a1.c &&
> > +	git add a1.c &&
> > +	git commit -m "5 lines" a1.c &&
> 
> It would be nice to use test_tick or test_commit for a more realistic
> history (with time marching forward).

As far as this test is concerned, that realism is not necessary, and comes
at the cost of readability.

> > +	sed s/3/3QaQb/ <a1.c | q_to_lf >tmp &&
> > +	mv tmp a1.c &&
> > +	git commit -m "2 more lines" a1.c &&
> 
> It's probably just me, but the bit with Q makes it hard for me to
> follow.  Maybe there's a simpler way?

Maybe. I did not find it, else I would have used it.

> "sed -e '3aa' -e '3ab'" works here, but I don't know how portable it
> is.

My experience with BSD sed and the `a` command made me highly suspicious,
that's why I did not go down that route.

Besides, that `sed` invocation does not really look intuitive either.

> I'd be more tempted to do
> 
> 	test_write_lines 1 2 3 4 5 >a1.c &&
> 	...
> 
> 	test_write_lines 1 2 3 a b 4 5 >a1.c &&
> 	...
> 
> 	test_write_lines 1 2 3 a b c 4 5 >a1.c &&
> 	...
> 
> which is concise and has obvious behavior.

That works for me!

Ciao,
Dscho