[kwlug-disc] bash strict mode (set -e)? - not so fast unless you know these quirks

[Paul Nijjar]

This got caught by the autoreject (probably because Hugh is a GTALUG
member, not a KWLUG one).

----- Forwarded message from "D. Hugh Redelmeier" <hugh at mimosa.com> -----

Date: Thu, 14 Aug 2025 16:24:57 -0400 (EDT)
From: "D. Hugh Redelmeier" <hugh at mimosa.com>
To: Ron via Talk <talk at lists.gtalug.org>
cc: KWLUG <kwlug-disc at kwlug.org>, sf-lug at linuxmafia.com, Ron <ron at bclug.ca>
Subject: Re: [GTALUG] bash strict mode (set -e)? - not so fast unless you
 know these quirks

> From: Ron via Talk <talk at lists.gtalug.org>

> Quick question:
> 
> * how regularly do you put `set` in your bash scripts?

As often as I remember to.

> * how often do you trap signals in your bash scripts?

Only ones that I put in "production" use.  It's often a sign to switch 
languages.  Generally: to clean up resources when a failure happens.

> On to the video:
> 
> > The Problem with Bash 'Strict Mode'
> >
> > - Why I Don’t Use `set -euo pipefail`
> 
> 
> More bash enlightenment by Dave @ YSAP on the topic of bash strict mode (set
> -e).
> 
> 
> There are some damned weird gotchas to beware of! I had no idea.
> 
> 
> His "contrived" example of how it can make your life much harder:
> 
> $ ((i=0)) ; echo "i: ${i} exit code: $?"
> i: 0 exit code: 1

That's what ((expression)) is supposed to do.

	If the value of the expression is non-zero, the return status is 
	0; otherwise the return status is 1.

I don't use ((expression)).  I don't doubt that it is sometimes convenient

The idea of assignment being legal within an ((expression)) does violence 
to the original model of the Bourne Shell.  As do the rules for 
double-quotes.

> 
> 
> Longer form:
> 
> $ ((i=-1)) ; echo $?
> 0
> $ ((i=0)) ; echo $?
> 1
> $ ((i=1)) ; echo $?
> 0

Look at the part of the man page that I quoted above.

> So, it's returning an "error" code of 1 that will cause `set -e` to bring your
> program to a grinding halt where it is inappropriate to do so. No errors were
> generated in that code.

What is the definition of "error" in bash?  A non-zero return status from 
a command.  So ((0)) is an error.

> He then covers `set -o pipefail` and again, weird behaviour.

I don't use it.  A case analysis is needed to know whether that signal is 
actually an error.


----- End forwarded message -----