Re: RFE: allow double quotes to group multi-words & be treated as 1 arg, w/DQ's stripped off -- including RH =~

[Greg Wooledge]

On Wed, Aug 04, 2010 at 02:02:39AM -0700, Linda Walsh wrote:
> Other string operators?  What other ones operate this way?
> 
> This is what I'm referring to:
> 
> a="hi"
> if [[ "hi" == "$a" ]]; then echo "this matches"; fi
> if [[ 'hi' == '$a' ]] then echo "this doesn't"; fi

You are misunderstanding the role of quotes.  Double quotes and single
quotes have some things in common:

 * They cause the content inside of them to be treated as a single word,
   instead of potentially multiple words.

 * They cause shell/pattern metacharacters such as space, asterisk,
   parentheses, braces, brackets, etc. to be treated as literal.

And one thing different:

 * Double quotes permit `...` and $ substitutions to occur.  Single do not.

In your examples, '$a' is matched as the literal string "dollar-sign a".
"$a" is matched as a *string* (not a pattern) whose value is the content
of the variable a.

The example you did NOT show is:

  if [[ "hi" == $a ]]; then ...

In that one, the $a is matched as a *pattern* (not a string) whose value
is the content of the variable a.  That is the significance of the double
quotes, or their lack.

In action:

  imadev:~$ a='*a'
  imadev:~$ if [[ "baa" = "$a" ]]; then echo yes; else echo no; fi
  no
  imadev:~$ if [[ "baa" = $a ]]; then echo yes; else echo no; fi
  yes

> I would prefer this work:
> 
> a="h."
> 
> if [[ "hi" =~ "$a" ]];....

Quoting the right hand side makes bash treat the contents of a as a
string rather than a pattern.  If you want the contents treated as a
pattern (or in this case, an ERE) then remove the quotes.

  a='^h.$'
  if [[ "hi" =~ $a ]]; ...

In fact, putting the ERE-pattern you want to match against into a variable
and then using =~ $variable *is* the recommended practice, if you need to
use =~ at all.  It is the only way to get consistent results among all
the different releases of bash that have =~, since the behavior changed
multiple times during the 3.x series.

In this particular case, I would use a glob rather than an ERE, because
"h?" is much simpler than "^h.$".