Message-ID: <20251021095946.3c4071fd@pumpkin>
Date: Tue, 21 Oct 2025 09:59:46 +0100
From: David Laight <david.laight.linux@...il.com>
To: Kevin Locke <kevin@...inlocke.name>
Cc: Jonathan Corbet <corbet@....net>, Randy Dunlap <rdunlap@...radead.org>,
 Thorsten Leemhuis <linux@...mhuis.info>, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] tools: remove unnecessary x suffix in test strings

On Mon, 20 Oct 2025 14:18:32 -0600
Kevin Locke <kevin@...inlocke.name> wrote:

> On Sun, 2025-10-19 at 11:17 +0100, David Laight wrote:
> > On Fri, 17 Oct 2025 16:28:12 -0600 Kevin Locke <kevin@...inlocke.name> wrote:  
> >> On Fri, 2025-10-17 at 15:12 +0100, David Laight wrote:  
> >>> On Thu, 16 Oct 2025 17:47:09 -0600 Kevin Locke <kevin@...inlocke.name> wrote:    
> >>>> Remove the "x" suffixes which unnecessarily complicate the code.    
> >>> 
> >>> The problems arise when $1 is (say) "-x", a simple LR parser will treat
> >>> [ -x = -x ] as a check for the file "=" being executable and then give
> >>> a syntax error for the second -x.
> >>> I can't imagine why shellcheck should warn about a leading x (or any other
> >>> character) provided field splitting is disabled (eg by "").
> >>> The leading x has definitely been needed in the past.    
> >> 
> >> Yep, it definitely has been.  The rationale on the wiki is that it's
> >> not necessary for modern shells (and presumably that it unnecessarily
> >> complicates the code): https://www.shellcheck.net/wiki/SC2268
> >> However, it notes Zsh had issues as recently as 2015, which is not as
> >> old as I would have expected.  
> > 
> > It doesn't really make much difference to the shell.
> > I really doubt you'll notice any difference in the time it takes to run.  
> 
> I agree.  However, I'm more concerned about readability and
> understandability for developers less familiar with the quirks of old
> shells.
> 
> >>> POSIX does require the three argument 'test' look for the middle argument
> >>> being an operator - but there might be historic shells that don't so that.
> >>> OTOH you are probably looking for code from the early 1980s!
> >>> But the POSIX spec (last time I read it) does point out the problems
> >>> with arbitrary strings being treated as operators causing complex expressions
> >>> be mis-parsed - which a leading x fixes.    
> >> 
> >> Good point.  I just reread it and can confirm that the current version
> >> still notes issues mitigated by the X prefix with "historical shells"
> >> and with greater than 4 argument cases:
> >> https://pubs.opengroup.org/onlinepubs/9799919799/utilities/test.html  
> > 
> > The fact that the 'greater than 4 argument case' can still require
> > a prefix character might be considered enough to make adding one all the
> > time 'good practise' even though it (probably) isn't actually needed.  
> 
> That seems reasonable to me, although I'd prefer omitting x and
> prohibiting >3 argument cases, which appears to be the route
> shellcheck takes with SC2268 + SC2166.

Ugg.
I know the parser is 'problematic' but you need -o (and -a) to get
moderately efficient expression evaluation.
If shellcheck objects to those I'd guess it also objects to ( and ).

You really don't want to use [ ... ] && [ ... ] because it goes right
out to the command pipeline parser.
Not to mention the lack of grouping for a || b && c


> > While I wouldn't error not having a prefix, generating an error when
> > there is one seems wrong.
> > What does shellcheck do with [ "$a" = "$b" -o "$c" = "$d" ] ?  
> 
> It only produces SC2166 (discouraging -o).  However, for 
> [ "x$a" = "x$b" -o "x$c" = "x$d" ] it also produces SC2268.
> 
> > Or even [ "$a" "$b" "$c" "$d" "$e" "$f "$g" ] ??  
> 
> This, and [ "$a" "$b" "$c" ] and [ "$a" "$b" ] produce parser error
> SC1073.  Unfortunately, this appears to be a long-standing shellcheck
> issue:  https://github.com/koalaman/shellcheck/issues/1645
> 
> >> I find && and || more readable, but I'm open to changing it if you
> >> feel strongly.  
> > 
> > They get parsed entirely differently and are likely to be measurably slower.
> > Just FYI I tend not to use 'if' statements at all, just (eg):
> > 	[ a = b ] && echo a == b
> >   
> >> Do I understand correctly that you are in favor of using the x prefix?
> >> I have a slight preference for leaving it off, but I'm open to adding
> >> it if you (or others) feel strongly.  
> > 
> > I wouldn't take them out and consider shellcheck wrong, but the suffix
> > were just stupid.  
> 
> Are you opposed to the patch I posted removing the suffixes?  I had
> tagged you as Suggested-by due to misreading your first post.  If the
> change is not something you'd suggest, I can repost without it.

The suffixes are just wrong.
If the shell treats the first parameter to [ as an operator and $1 is "-"
it processes [ -x = ... and looks for a file "=".
Without the suffix the same happens when $1 is "-x".

A conformant shell won't do this for a 3-argument [.
But I also suspect that any conformant shell supports -o and -a.
The 7-argument [ definitely needs the prefix to protect against unexpected
operators.

So I still think shellcheck is just wrong here.
It ought to be checking FOR a prefix when there are 4 or more arguments.
It is one of those idioms you have to get used to.

But at the end of the day it is probably your call.

	David

> 
> Thanks,
> Kevin

Powered by blists - more mailing lists