Re: bug in sh, probably, with test case

[Richard Hansen <rhansen%bbn.com@localhost>]

On 2012-12-20 14:22, Greg Troxel wrote:
> One of the git tests fails.  I have reduced the failure to a small test
> case:

I too have noticed this bug.  This is the test case I used:

  sh -c 'a="x  x"; unset b; printf %s\\n "${b-${a}}"'

Notice the two spaces between the two exes (to confirm that it's not 
doing field splitting).

It should print:

x  x

but on NetBSD it prints:

x
x

-Richard


> ----------------------------------------
> #!/bin/sh
>
> func() {
>      n=$1; shift
>      echo "$n      1 <$1>"
>      echo "$n      2 <$2>"
> }
>
> p="--foo --bar"
>
> func bare $p
> func quoted "$p"
> func alt "alt ${p:+$p}"
> ----------------------------------------
>
> I get results (netbsd-5, i386)
>
> sh:
> ----------------------------------------
> bare    1 <--foo>
> bare    2 <--bar>
> quoted  1 <--foo --bar>
> quoted  2 <>
> alt     1 <alt --foo>
> alt     2 <--bar>
> ----------------------------------------
>
> bash:
> ----------------------------------------
> bare    1 <--foo>
> bare    2 <--bar>
> quoted  1 <--foo --bar>
> quoted  2 <>
> alt     1 <alt --foo --bar>
> alt     2 <>
> ----------------------------------------
>
> ksh:
> ----------------------------------------
> bare    1 <--foo>
> bare    2 <--bar>
> quoted  1 <--foo --bar>
> quoted  2 <>
> alt     1 <alt --foo --bar>
> alt     2 <>
> ----------------------------------------
>
> and obviously the Linux shells do not split the contents of $p, because
> the git tests must pass there.
>
>
> POSIX requires that field splitting not be done on the results of
> substitutions in "":
>
>    
> http://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html#tag_18_06_02
>
> so I think our sh is wrong.
>
> (git's test is constructing a human-readable string explaining the
> arguments passed to the test, so it is doing essentially like the 'alt'
> line above, but with extra text: "....is valid${2:+ with options $2}".)