patchgit-subtree: Avoid using echo -n even indirectly

Since say uses echo, this uses echo -n, which is not portable - see 19c3c5fdcb35b66b792534c5dc4e8d87a3952d2a.

Without this commit, the output looks like:

... -n 1891/ 1936 (1883) -n 1892/ 1936 (1884) -n 1893/ 1936 (1885) ...

Signed-off-by: Paolo G. Giarrusso <p.giarrusso@gmail.com>
---

Please CC me on replies, as I am not subscribed to this mailing list. I am tracking this submission via https://github.com/git/git/pull/61, which I'll duly close myself when the discussion is resolved.

 contrib/subtree/git-subtree.sh | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

On Wed, Oct 9, 2013 at 11:57 AM, Paolo G. Giarrusso <p.giarrusso@gmail.com> wrote:

Reviewers might wish to know that "say" in git-subtree is defined as

        say()
        {
                if [ -z "$quiet" ]; then
                        echo "$@" >&2
               fi
        }

Hence the "if" and the redirect.

(Resending without HTML).

On Wed, Oct 9, 2013 at 12:20 PM, Tay Ray Chuan <rctay89@gmail.com> wrote:

An additional note for reviewers and appliers: the original and the patched codeboth embed a literal ^M, not a new line, go to back to the beginning of the line and overwrite it, so the above is not a consequence of line-wrap.

I used git-format-patch and git-send-email, and the ^M is visible in Vim in the exported patch (that's why I didn't remark on it). Seeing the email, I wonder whether there's hope something like that can be preserved in an email, and whether the code should use some escape sequence instead.

Indeed. I considered having a variant of `say` instead of inlining and customizing it, but for once I decided to keep this simple, since this variant of `say` is currently used only once. Otherwise, one could change say to use printf, but that's more invasive.

Cheers,

Am 10/9/2013 12:32, schrieb Paolo Giarrusso:

As you are using printf, you can easily do:

			printf '%s\r' "..."

without using ^M.

-- Hannes

Paolo Giarrusso <p.giarrusso@gmail.com> writes:

"invasive" in the sense that it impacts indirectly more callers, but are there really cases where "echo" is needed when calling "say"? Aren't there other potential bugs when arbitrary strings are passed to "say", that would be fixed by using printf once and for all?

The patch would look like the one I did in 89b0230a20 (Wed Aug 7 2013, die_with_status: use "printf '%s\n'", not "echo").

On Wed, Oct 9, 2013 at 1:26 PM, Matthieu Moy <Matthieu.Moy@grenoble-inp.fr> wrote:

(1) Changing the implementation of say to use printf "%s\n" would be trivial, and I think would address your concerns.

But I was concerned about code duplication; one could additionally make say reusable in this single call site, instead of inlining and customizing it by replacing the "\n" with "\r". But for that, you need to either (2) add an explicit \n to all callers (invasive & error prone), or (3) make `say` parse the `-n` option and conditionally add "\n" to the format string or to a final argument, if -n is not specified; this would affect no current caller, but complicate the implementation of say. Doing that for just one call site has too much potential for breakage, so I'm not sure I'd do it. (I'm not even sure on what should `say` do when `-n` is not the first argument).

Options (1), (2) and (3) are mutually alternative; my favorite is (1).

I can see your points about opportunity, especially after looking at the commit message of the patch of yours you linked.

I see your point. But note that using printf like in die_with_status after that commit wouldn't be reusable here in all call sites, because it always prints a newline.

Cheers,

On Wed, Oct 09, 2013 at 02:03:24PM +0200, Paolo Giarrusso wrote:

Yeah, I think we should do that. I had the same thought as Matthieu when I read your initial patch; there are real portability bugs caused by using "echo" that would be fixed.

However, that is somewhat orthogonal to the bug you are fixing. For handling this one site, I think it would be fine to just convert it to use printf, as your patch did. As you noted, the alternatives:

...are either annoying or complicated (and in particular, parsing "-n" means that callers need to be aware that 'say "$foo"' might accidentally trigger "-n" if $foo comes from the user). So the sanest interface is probably "say_nonl" or something similar. But since there would only be one caller, I don't see much point in factoring it out.

Me too. :)

-Peff

Paolo Giarrusso wrote:

Yes, please.  Mind if I amend it to

	printf "%s\r" "$revcount/$revmax ($createcount)" >&2

?

[...]

I agree with the other reviewers that this should be fixed to use printf, too, but that's another topic.

Thanks, Jonathan

On Wed, Oct 9, 2013 at 11:11 PM, Jonathan Nieder <jrnieder@gmail.com> wrote:

Please do go ahead, by all means (arguably as a different commit, but those are minor details).

Seconded.