Re: "grep '\]'" warnings suggest a Gnulib DFA patch

[Paul Eggert]

On 6/3/22 20:08, Bruno Haible wrote:

> But when I think about the thousands of people who use regular expressions
> out there. How would they remember that in parentheses both should be
> backslash-escaped in EREs
>      \(   \)

It's even weirder, in that POSIX says unmatched ')' is treated like '\)' 
in an ERE, which is why gnulib/lib/dfa.c does not warn about it.


> but brackets and braces are asymmetric
>      \[   ]
>      \{   }

Thanks, good catch about \}. We should treat it like \]. (And this means 
regex-quote.c is buggy in a different way, sigh....)


> Even if the warning message you install in grep has 3 or 5 lines and goes
> into all details, we are not serving the community if we force them to adhere
> to asymmetric rules, where up to now they could use symmetric rules.

Thanks, and I see Jim agrees too. I installed the first two attached 
patches into Gnulib to do that and to fix regex-quote, propagated this 
into Grep, and installed the last attached patch to Grep to document this.

At some point the behavior of \], \}, and all the other stuff the Grep 
manuals new "Problematic Expressions" node should be documented in 
gnulib/doc/regex.texi too. I'll cc this to Reuben to see whether he has 
the time.

It might be useful for GNU grep to have a --pedantic flag to check 
regular expression portability, to reject unportable REs like '\]'. But 
any such feature can wait until after the next GNU grep release.

0001-dfa-do-not-warn-about-and.patch
Description: Text Data

0002-regex-quote-in-EREs.patch
Description: Text Data

0001-doc-document-and.patch
Description: Text Data