Re: [PATCH] t2017: redo physical reflog existance check

[Erick Mattos <erick.mattos@gmail.com>]

2010/7/22 Junio C Hamano <gitster@pobox.com>:
> Erick Mattos <erick.mattos@gmail.com> writes:
>
>> To make the new orphan branch ready to have a reflog on that config
>> was as simple as creating a "touch file" for reflog.  This is the goal
>> achieved by code.
>
> I have to say that you are somewhat confused about the _goal_ then.  touch
> is not a goal, it is means to a goal.

I believe you have understood the whole idea of my previous email even
though _words_ means a lot to you.  :-1

I meant that that was the code goal.  It was the implementation
objective which by itself was the solution chosen for the real
problem.

> It does not matter how you implement the user visible effect, be it a
> creation of an empty file, or some other means [*1*].  What matters is
> that the user won't get a reflog for a branch that really didn't get
> created and must-fail "rev-parse --verify" test checks that.
>
> Another thing that could matter would be that future actions that want to
> create a reflog for the same branch (perhaps after the user switches to
> 'master', another attempt is made to create eta with "checkout -b eta") or
> another branch with a similar or related name (say "eta/real") are not get
> broken by whatever you do to implement the "we want to create a reflog
> when a ref is actually made but not right now" feature.  Perhaps the right
> way to test that would be to actually try to run such operations and make
> sure they do not fail.

You were cutting off redundancy checks, weren't you?  If there is no
reflog file (tested by "test -f"), consequently no reflog (tested by
"rev-parse --verify") and no branch with the chosen name, certainly
someone can create one the "almost used" name, once there is no any
difference from a normal situation.

> [Footnote]
>
> *1* For example, you could have implemented the feature by adding a config
> item in ".git/config: [branch "eta"] need-to-create-reflog", and taught
> refs.c::update_ref() to pay attention to it (I am not saying that it would
> be a better implementation).

About the late parenthesis: thank God! ;-)

I don't see a need for so much reluctance: "test -f" is not a taboo
inside a script in t folder and the added tests don't change anything
about the design and implementation which IMHO is well fit.

With those two patch lines of mine "--orphan with -l and
core.logAllRefUpdates=false" testing script is finished.

Finally: you are the man in charge so I would really like to enforce
that if you need me to do anything more I will be _really_ glad to
help.  I love git and everything good done to it it is done to me too
as one of its daily user.

Best regards