Re: git-bisect: weird usage of read(1)

["Reece Dunn" <msclrhd@googlemail.com>]

2008/8/11 Francis Moreau <francis.moro@gmail.com>:
> Hello
>
> On Mon, Aug 11, 2008 at 4:15 PM, Johannes Schindelin
> <Johannes.Schindelin@gmx.de> wrote:
>> Hi,
>>
>> On Mon, 11 Aug 2008, Francis Moreau wrote:
>>
>>> I found this in git bisect:
>>>
>>>               printf >&2 'Are you sure [Y/n]? '
>>>               case "$(read yesno)" in [Nn]*) exit 1 ;; esac
>>>
>>> which looks very weird since read(1) returns a status and not the
>>> string reads from std input.
>>>
>
> sorry I should have said that there's a status but no output...
>
>>> Am I missing something ?
>>
>> Yes.  "$()" does not return the status, but the output.
>>
>
> But what's the output in that case ?

Using cygwin+bash, I get:

$ echo $(read yesno)
n

$ echo $(read yesno; echo $yesno)
n
n
$ $(read yesno) && echo yes || echo no
n
yes
$ $(read yesno) && echo yes || echo no
y
yes
$ case "$(read yesno)" in [Nn]*) echo "no" ;; esac
n
$ case "$(read yesno)" in [Nn]*) echo "no" ;; esac
y
$ case "$(read yesno; echo $yesno)" in [Nn]*) echo "no" ;; esac
n
no
$ case "$(read yesno; echo $yesno)" in [Nn]*) echo "no" ;; esac
y

So

>>>               case "$(read yesno)" in [Nn]*) exit 1 ;; esac

does not work as expected. Replacing this with

               case "$(read yesno; echo $yesno)" in [Nn]*) exit 1 ;; esac

would work as intended, as Mikael has pointed out.

- Reece