From: Johannes Sixt <j.sixt@viscovery.net>
Date: Mon, 09 May 2011 06:39:56 GMT
Subject: Re: [PATCH] t1507: change quoting in test_did_you_mean to a more general one
Message-ID: <4DC78C3C.6060004@viscovery.net>
In-Reply-To: <7voc3cqyfp.fsf@alter.siamese.dyndns.org>

Am 5/9/2011 6:40, schrieb Junio C Hamano:
> Johannes Sixt <j.sixt@viscovery.net> writes:
> 
>> - The behavior of quoting at the right of :- when the ${...:-...} exansion
>> appears in double-quotes was debated recently at length at the Austin
>> group (which revises the POSIX standard). You better move the expansions
>> to assignments of temporary variables,...
> 
> What alternative semantics do Austin folks have in mind, by the way?  Just
> declare this undefined?

Most of the debate centered around how to quote a closing brace: the
problem is that usually the backslash within double-quotes is only special
when followed by $, `, or \, therefore, it should not be possible to quote
a closing brace using \}. It turned out that different shells handled this
particular case differently. IIRC, there were also discrepancies whether
whitespace generated on the RHS of :- (or other operators) was retained or
discarded. The conclusion for us as application developers is not to
depend on too many subtle details in such a variable expansions because
the specification is too vague, shells implement it differently, and some
even get it outright wrong (as the case discovered by Kascper).

> -	printf "fatal: Path '$2$3' $4, but not ${5:-'$3'}.\n" >expected &&
> -	printf "Did you mean '$1:$2$3'${2:+ aka '$1:./$3'}?\n" >>expected &&
> +	sq="'"

This cuts the && chain.

> +	cat >expected <<-EOF &&
> +	fatal: Path '$2$3' $4, but not ${5:-$sq$3$sq}.
> +	Did you mean '$1:$2$3'${2:+ aka $sq$1:./$3$sq}?
> +	EOF

Looks good otherwise; I tested this with various shells, and there were no
surprises.

-- Hannes