[BUGS] test failure in t4014-format-patch.sh with master

[BUGS] test failure in t4014-format-patch.sh with master

[Remi Vanicat]

Hello,

While trying to wrote a test for git, I discover that there is a test
failure in master on my computer:

$ ./t4014-format-patch.sh
[....]
* FAIL 8: replay did not screw up the log message

 git cat-file commit rebuild-1 | grep "^Side .* with .* backslash-n"


It seem that now, somewhere a \n has been interpreted as a newline.

Bisecting point the problem to the 2e6e3e82 commit "git-am: cope
better with an empty Subject: line"


-- 
Rémi Vanicat

Re: [BUGS] test failure in t4014-format-patch.sh with master

[Linus Torvalds]

On Thu, 17 Apr 2008, Remi Vanicat wrote:
> 
> While trying to wrote a test for git, I discover that there is a test
> failure in master on my computer:

Hmm. Doesn't fail for me.

> $ ./t4014-format-patch.sh
> [....]
> * FAIL 8: replay did not screw up the log message
> 
>  git cat-file commit rebuild-1 | grep "^Side .* with .* backslash-n"
> 
> It seem that now, somewhere a \n has been interpreted as a newline.

I wonder if it is some shell-specific variable expansion issue. What's 
your /bin/sh? (I assume, from your email address, that you're using 
Debian, and I thought Debian used bash, but maybe that's not true).

What does

	a="hello\nhi"
	echo $a

result in?

		Linus

Re: [BUGS] test failure in t4014-format-patch.sh with master

[Björn Steinbrink]

On 2008.04.17 08:11:52 -0700, Linus Torvalds wrote:
> 
> 
> On Thu, 17 Apr 2008, Remi Vanicat wrote:
> > 
> > While trying to wrote a test for git, I discover that there is a test
> > failure in master on my computer:
> 
> Hmm. Doesn't fail for me.
> 
> > $ ./t4014-format-patch.sh
> > [....]
> > * FAIL 8: replay did not screw up the log message
> > 
> >  git cat-file commit rebuild-1 | grep "^Side .* with .* backslash-n"
> > 
> > It seem that now, somewhere a \n has been interpreted as a newline.
> 
> I wonder if it is some shell-specific variable expansion issue. What's 
> your /bin/sh? (I assume, from your email address, that you're using 
> Debian, and I thought Debian used bash, but maybe that's not true).
> 
> What does
> 
> 	a="hello\nhi"
> 	echo $a
> 
> result in?

Debian has its own dash which is supposed to be a lightweight
alternative to bash and "for checking POSIX compliance of scripts"
(quote from the packages description). I don't happen to know off-hand
whether POSIX says that echo should default to -e, but dash seems to do
that:

$ dash
$ a="hello\nhi"
$ echo $a
hello
hi

Björn

Re: [BUGS] test failure in t4014-format-patch.sh with master

[Rémi Vanicat]

2008/4/17, Linus Torvalds <torvalds@linux-foundation.org>:
>
>
>  On Thu, 17 Apr 2008, Remi Vanicat wrote:
[...]
>  > $ ./t4014-format-patch.sh
>  > [....]
>  > * FAIL 8: replay did not screw up the log message
>
>  Hmm. Doesn't fail for me.

[...]

>  I wonder if it is some shell-specific variable expansion issue. What's
>  your /bin/sh? (I assume, from your email address, that you're using
>  Debian, and I thought Debian used bash, but maybe that's not true).

Debian used bash, but I use dash as /bin/sh

>  What does
>
>         a="hello\nhi"
>         echo $a
>
>  result in?

here is the problem:

$/bin/sh
$ a="hello\nhi"
$ echo $a
hello
hi

Mmm, I Wonder what posix said about this, but it look like a bug with dash

Re: [BUGS] test failure in t4014-format-patch.sh with master

[Linus Torvalds]

On Thu, 17 Apr 2008, Björn Steinbrink wrote:
> 
> Debian has its own dash which is supposed to be a lightweight
> alternative to bash and "for checking POSIX compliance of scripts"
> (quote from the packages description). I don't happen to know off-hand
> whether POSIX says that echo should default to -e, but dash seems to do
> that:
> 
> $ dash
> $ a="hello\nhi"
> $ echo $a
> hello
> hi

Ahh. So that "echo" should just be replaced with a 'printf "%s\n"' 
instead? 

We have a _lot_ of "echo"s though. I suspect the only ones we'd ever catch 
are the ones explicitly tested for. I suspect that the dash echo is just 
broken.

		Linus

Re: [BUGS] test failure in t4014-format-patch.sh with master

[Björn Steinbrink]

On 2008.04.17 08:44:07 -0700, Linus Torvalds wrote:
> 
> 
> On Thu, 17 Apr 2008, Björn Steinbrink wrote:
> > 
> > Debian has its own dash which is supposed to be a lightweight
> > alternative to bash and "for checking POSIX compliance of scripts"
> > (quote from the packages description). I don't happen to know off-hand
> > whether POSIX says that echo should default to -e, but dash seems to do
> > that:
> > 
> > $ dash
> > $ a="hello\nhi"
> > $ echo $a
> > hello
> > hi
> 
> Ahh. So that "echo" should just be replaced with a 'printf "%s\n"' 
> instead? 
> 
> We have a _lot_ of "echo"s though. I suspect the only ones we'd ever catch 
> are the ones explicitly tested for. I suspect that the dash echo is just 
> broken.

Wow, seems that "echo" in itself is "broken". According to POSIX[1]
backslashes induce implementation defined behaviour, and there seem to
be two historic versions of echo, one that knows about -n and one that
interprets escape sequences. So echo is totally non-portable unless you
avoid -n as well as backslashes. Ouch.

The POSIX docs, as well as random Google results, indeed recommend to
use printf instead. But that switch seems painful as well...

Björn

[1] http://www.opengroup.org/onlinepubs/009695399/utilities/echo.html

Re: [BUGS] test failure in t4014-format-patch.sh with master

[Jeff King]

On Thu, Apr 17, 2008 at 08:44:07AM -0700, Linus Torvalds wrote:

> Ahh. So that "echo" should just be replaced with a 'printf "%s\n"' 
> instead? 
> 
> We have a _lot_ of "echo"s though. I suspect the only ones we'd ever catch 
> are the ones explicitly tested for. I suspect that the dash echo is just 
> broken.

As Bjorn mentioned, it's POSIX.  Junio and I already hunted down quite a
few of these, so I think we are OK in general, and this is just a new
one that got introduced.

-Peff

Re: [BUGS] test failure in t4014-format-patch.sh with master

[Jeff King]

On Thu, Apr 17, 2008 at 12:24:22PM -0400, Jeff King wrote:

> As Bjorn mentioned, it's POSIX.  Junio and I already hunted down quite a
> few of these, so I think we are OK in general, and this is just a new
> one that got introduced.

Oops, I meant to mention this thread:

  http://article.gmane.org/gmane.comp.version-control.git/48419

-Peff

Re: [BUGS] test failure in t4014-format-patch.sh with master

[Junio C Hamano]

"Rémi Vanicat" <vanicat@debian.org> writes:

> 2008/4/17, Linus Torvalds <torvalds@linux-foundation.org>:
>>
>>  On Thu, 17 Apr 2008, Remi Vanicat wrote:
> [...]
>>  > $ ./t4014-format-patch.sh
>>  > [....]
>>  > * FAIL 8: replay did not screw up the log message
>>
>>  Hmm. Doesn't fail for me.
>
> [...]
>
>>  I wonder if it is some shell-specific variable expansion issue. What's
>>  your /bin/sh? (I assume, from your email address, that you're using
>>  Debian, and I thought Debian used bash, but maybe that's not true).
>
> Debian used bash, but I use dash as /bin/sh
>
>>  What does
>>
>>         a="hello\nhi"
>>         echo $a
>>
>>  result in?

So perhaps this one is the troublemaker?

 git-am.sh |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/git-am.sh b/git-am.sh
index 646dc45..5b20bbd 100755
--- a/git-am.sh
+++ b/git-am.sh
@@ -330,7 +330,7 @@ do
 		SUBJECT="$(sed -n '/^Subject/ s/Subject: //p' "$dotest/info")"
 		case "$keep_subject" in -k)  SUBJECT="[PATCH] $SUBJECT" ;; esac
 
-		(echo "$SUBJECT" ; echo ; cat "$dotest/msg") |
+		(printf '%s\n' "$SUBJECT" ; echo ; cat "$dotest/msg") |
 			git stripspace > "$dotest/msg-clean"
 		;;
 	esac

Re: [BUGS] test failure in t4014-format-patch.sh with master

[Linus Torvalds]

On Fri, 18 Apr 2008, Junio C Hamano wrote:
>
> -		(echo "$SUBJECT" ; echo ; cat "$dotest/msg") |
> +		(printf '%s\n' "$SUBJECT" ; echo ; cat "$dotest/msg") |

Well, get rid of the superfluous second 'echo' then too, replacing it 
with another '\n'. IOW:

	( printf '%s\n\n' "$SUBJECT" ; cat "$dotest/msg" ) | 

instead.

		Linus

Re: [BUGS] test failure in t4014-format-patch.sh with master

[Rémi Vanicat]

2008/4/18, Linus Torvalds <torvalds@linux-foundation.org>:
>
>
>  On Fri, 18 Apr 2008, Junio C Hamano wrote:
>  >
>  > -             (echo "$SUBJECT" ; echo ; cat "$dotest/msg") |
>  > +             (printf '%s\n' "$SUBJECT" ; echo ; cat "$dotest/msg") |
>
>
> Well, get rid of the superfluous second 'echo' then too, replacing it
>  with another '\n'. IOW:
>
>         ( printf '%s\n\n' "$SUBJECT" ; cat "$dotest/msg" ) |
>
>  instead.

This work for me: no more test failure on my computer

Re: [BUGS] test failure in t4014-format-patch.sh with master

[Junio C Hamano]

"Rémi Vanicat" <vanicat@debian.org> writes:

> 2008/4/18, Linus Torvalds <torvalds@linux-foundation.org>:
>>
>>
>>  On Fri, 18 Apr 2008, Junio C Hamano wrote:
>>  >
>>  > -             (echo "$SUBJECT" ; echo ; cat "$dotest/msg") |
>>  > +             (printf '%s\n' "$SUBJECT" ; echo ; cat "$dotest/msg") |
>>
>>
>> Well, get rid of the superfluous second 'echo' then too, replacing it
>>  with another '\n'. IOW:
>>
>>         ( printf '%s\n\n' "$SUBJECT" ; cat "$dotest/msg" ) |
>>
>>  instead.
>
> This work for me: no more test failure on my computer

Thanks for confirmation.  I reproduced the breakage and the fix myself
with "make SHELL_PATH=/bin/dash" myself, too.