From: SZEDER Gábor <szeder.dev@gmail.com>
Date: Sat, 24 Oct 2026 19:43:09 GMT
Subject: Re: [PATCH v1 1/1] t1450: use mv -f when overwriting a read only file
Message-ID: <at0KTewYGKuUTAUJ@szeder.dev>
In-Reply-To: <20261024183056.6342-1-tboegi@web.de>

On Sat, Oct 24, 2026 at 08:30:56PM +0200, tboegi@web.de wrote:
> From: Torsten Bögershausen <tboegi@web.de>
> 
> test case 38 in t/t1450-fsck.sh replaces a loose blob, "$tmp" below,
> with a trucated version "obj.tmp" using
> 
> mv obj.tmp "$obj"
> 
> This fails on some sytems, like older MacOS.
> `man mv`
> includes this information:
>   The following options are available:
>   -f      Do not prompt for confirmation before overwriting the destination
>            path.  (The -f option overrides any previous -i or -n options.)
> and further down:
>   If the destination path does not have a mode which permits writing, mv
>   prompts the user for confirmation as specified for the -i option.
> 
> Running the test case in debug and verbose mode `./t1450-fsck.sh -d -v`
> shows this lines (which I have shortened to fit into 72 characters)
> 
> override r--r--r--  user/staff for .git/objects/fc/a75a5be3823dc0560da5?
> (y/n [n]) not overwritten
> fca75a5be3823dc0560da57ade678f212aeb33f6
> 
> POSIX mentions that the observed
> "prompt to standard error and read a line from standard input"
> should not be done when "standard input is a terminal".
> So we may blame the "standard input is a terminal" logic to fail.
> 
> Side note: Answering with 'y' to the question of `mv` like this
> (echo y | mv obj.tmp "$obj")
> makes the test case pass.
> 
> To make things not too complicated use the most portable form
> mv -f obj.tmp "$obj"
> 
> Signed-off-by: Torsten Bögershausen <tboegi@web.de>
> ---
> 
> So it was not the shell being old - but /bin/mv
> 
> t/t1450-fsck.sh | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/t/t1450-fsck.sh b/t/t1450-fsck.sh
> index 16cc5d5dcb..335dcbf0af 100755
> --- a/t/t1450-fsck.sh
> +++ b/t/t1450-fsck.sh
> @@ -548,7 +548,7 @@ test_expect_success 'rev-list --verify-objects with truncated loose blob' '
>  		# Truncate the loose blob such that its header can still be
>  		# parsed, but reading the object data fails mid-stream.
>  		test_copy_bytes 64 <"$obj" >obj.tmp &&
> -		mv obj.tmp "$obj" &&
> +		mv -v obj.tmp "$obj" &&

You want force, not verbosity.

Btw, don't you have a bit of a dejà vu? :)

https://lore.kernel.org/git/20240502185903.GA11895@tb-raspi4/T/#u

>  
>  		test_must_fail git rev-list --verify-objects "$blob" 2>err &&
>  		test_grep "hash mismatch" err
> -- 
> 2.50.0.rc0.46.g7014b55638
>