From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andrew Dunstan <andrew@dunslane.net>
cc: pgsql-committers@lists.postgresql.org
Subject: Re: pgsql: Check that we have a working tar before trying to use it
In-reply-to: <2e118a11-6936-21ec-6782-e9dbb20972c2@dunslane.net>
References: <E1muynv-0006GS-R4@gemulon.postgresql.org>
 <3655283.1638977975@sss.pgh.pa.us>
 <2e118a11-6936-21ec-6782-e9dbb20972c2@dunslane.net>
Comments: In-reply-to Andrew Dunstan <andrew@dunslane.net>
	message dated "Wed, 08 Dec 2021 10:56:46 -0500"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <3657696.1638980149.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Wed, 08 Dec 2021 11:15:49 -0500
Message-ID: <3657697.1638980149@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/3657697.1638980149%40sss.pgh.pa.us>
Precedence: bulk

Andrew Dunstan <andrew@dunslane.net> writes:
> On 12/8/21 10:39, Tom Lane wrote:
>> ISTM that effectively restricts the test to only running
>> on machines with GNU tar, which basically removes all the
>> interest of it.  We know what GNU tar does ... it's the
>> weird legacy tar versions that might teach us something.
>> See 57b5a9646 for a recent example of the sort of bug
>> this test can no longer find.

> I tested on a freebsd system before I did this for that reason. It's not
> using GNU tar:
> user@freebsd:~ $ tar --version
> bsdtar 3.5.1 - libarchive 3.5.1 zlib/1.2.11 liblzma/5.2.5 bz2lib/1.0.8

Right, but on AIX:

tgl@gcc119:[/home/tgl]tar --version
tar: Not a recognized flag: -
Usage: tar -{c|r|t|u|x} [ -BdDEFhilmopRUsvwZ ] [ -Number ] [ -f TarFil e ]
           [ -b Blocks ] [ -S [ Feet ] | [ Feet@Density ] | [ Blocksb ] ]
           [ -L InputList ] [-X ExcludeFile] [ -N Blocks ] [ -C Directory =
] File ...
Usage: tar {c|r|t|u|x} [ bBdDEfFhilLXmNopRsSUvwZ[0-9] ] ]
           [ Blocks ] [ TarFile ] [ InputList ] [ ExcludeFile ] =

           [ [ Feet ] | [ Feet@Density ] | [ Blocksb ] ] [-C Directory ] F=
ile ...
tgl@gcc119:[/home/tgl]echo $?
1

> Do you have an alternative test we could use?

I think you need to be straight up about it, say

	touch foo; tar cf foo.tar foo

(At least on the Unix machines I tried, it works to use /dev/null
as the output file, saving one cleanup step.  But I don't know
if that'll work on Windows.)

			regards, tom lane