Re: [PATCH 2/3] fast-export: add new --refspec option

Felipe Contreras [off-list ref] writes:

+test_expect_success 'use refspec' '
+	git fast-export --refspec refs/heads/master:refs/heads/foobar master | \
+		grep "^commit " | sort | uniq > actual &&

You do not need backslash after the pipe symbol at the end of line;
the shell knows you haven't finished speaking at that point.

The usual "pipe hides the error status of upstream commands" applies
here.  The command may die after writing enough to fill the pipe
buffer and showing the lines that begin with "commit".

Also it makes it harder to debug the test when something goes wrong.

By the way, don't you find that something does not feel quite right
with this command line?

    git fast-export --refspec=refs/heads/master:refs/heads/foobar master 

Why do we even have to say "master" at the end, when the other
option makes it clear that we are shipping "master" out?

Without thinking ramifications through, my gut feeling is that it
would feel more natural if we took:

    git fast-export master:foobar

to mean the same thing (which is what happens to the users of "git
push").  Is there a case where you have some ref on the left hand
side of the --refspec but you do not push out the history leading to
it?

With such an update, this part of the test would of course look
like:

	git fast-export master:foobar >actual.dump &&
        grep "^commit " actual.dump |
        sort -u >actual &&
        ...

and we do not need a new option.  Just a new extension to express
what gets pushed out.

But of course I may be missing some cases why there need to be a
separate option.

On Thu, May 9, 2013 at 5:32 PM, Junio C Hamano [off-list ref] wrote:

Felipe Contreras [off-list ref] writes:

quoted

+test_expect_success 'use refspec' '
+     git fast-export --refspec refs/heads/master:refs/heads/foobar master | \
+             grep "^commit " | sort | uniq > actual &&

You do not need backslash after the pipe symbol at the end of line;
the shell knows you haven't finished speaking at that point.

The usual "pipe hides the error status of upstream commands" applies
here.  The command may die after writing enough to fill the pipe
buffer and showing the lines that begin with "commit".

Also it makes it harder to debug the test when something goes wrong.

By the way, don't you find that something does not feel quite right
with this command line?

    git fast-export --refspec=refs/heads/master:refs/heads/foobar master

Why do we even have to say "master" at the end, when the other
option makes it clear that we are shipping "master" out?

Without thinking ramifications through, my gut feeling is that it
would feel more natural if we took:

    git fast-export master:foobar

to mean the same thing (which is what happens to the users of "git
push").  Is there a case where you have some ref on the left hand
side of the --refspec but you do not push out the history leading to
it?

With such an update, this part of the test would of course look
like:

        git fast-export master:foobar >actual.dump &&
        grep "^commit " actual.dump |
        sort -u >actual &&
        ...

and we do not need a new option.  Just a new extension to express
what gets pushed out.

But of course I may be missing some cases why there need to be a
separate option.

Of course, but how do you implement that? That's mixing refspecs and
revlist arguments, which AFAIK don't mix:

% git fast-export ^next:new-next master:new-master --not
refs/tags/*:refs/tags/backup/*

-- 
Felipe Contreras