Re: shell compatibility issues with SunOS 5.10

From: Brandon Casey <casey@nrlssc.navy.mil>
Date: May 6, 2009, 18:14 UTC
Message-ID: <-ElxRhvpfY_jx1Ps8nJ42rHdrKbR03T1y96WpGK19gM@cipher.nrlssc.navy.mil>
In-Reply-To: <20090506055913.GA9701@dektop>

Nguyen Thai Ngoc Duy wrote:

> Hi,
> 
> I did "make test" on a SunOS 5.10 and it failed.

Show 10 quoted lines

> diff --git a/t/t4200-rerere.sh b/t/t4200-rerere.sh
> index b68ab11..61ccdee 100755
> --- a/t/t4200-rerere.sh
> +++ b/t/t4200-rerere.sh
> @@ -57,7 +57,7 @@ test_expect_success 'conflicting merge' '
>  	test_must_fail git merge first
>  '
>  
> -sha1=$(sed -e 's/	.*//' .git/MERGE_RR)
> +sha1=$(cut -f 1 .git/MERGE_RR)

Are you using /bin/sed? I think it has a problem with tabs. /usr/xpg4/bin/sed works correctly, but it has a problem with files that are not newline terminated. So you will get other errors. I have a set of "ugly" patches that I will follow this email with that allow me to compile on Solaris 10 while skipping the following tests:

   GIT_SKIP_TESTS='
      t3900.2[23]
      t3901.[67]
      t6030.1[23]
      t8005.*
   '

The t3900 and t3901 are due to iconv failures.

The t6030 issues are due to a flaw in Sun's ksh, i.e. /usr/xpg4/bin/sh which I use for testing. This ksh seems to only call a trap that is set within at most the next outer-layer function. In other words, if func1 sets a trap, and calls func2 which calls func3 which exits, then the trap in func1 will _not_ be called. If instead, func2 exits, then the trap will be called. Bisect does such a thing. You should not have a problem with other ksh or bash.

I had assumed t8005 was failing because of iconv, but since you have pointed out the extended RE in grep, some of these should pass. Converting to egrep allows tests 1, 4 and 5 to pass for me. So my skip expression can be changed to t8005.[23].

>  rr=.git/rr-cache/$sha1
>  test_expect_success 'recorded preimage' "grep ^=======$ $rr/preimage"

Patches to follow.

-brandon