Bug #122794 for PPR: s///e RHS is not validated

Sun Aug 13 12:04:10 2017 arc [...] cpan.org - Ticket created

Subject:

s///e RHS is not validated

Perl refuses to compile programs with syntax errors on the right-hand side of an s///e: $ perl -ce 's/./3 4/e' Number found where operator expected at -e line 1, near "3 4" (Missing operator before 4?) syntax error at -e line 1, near "3 4" -e had compilation errors. but PPR happily accepts them; it apparently ignores the /e flag. There are some very tricky cases hiding here; in particular, the RHS of the substitution can contain heredocs and formats and so on.

Mon Aug 14 07:58:22 2017 damian [...] conway.org - Correspondence added

Subject:	Re: [rt.cpan.org #122794] s///e RHS is not validated
Date:	Mon, 14 Aug 2017 13:57:25 +0200
To:	bug-PPR [...] rt.cpan.org
From:	Damian Conway <damian [...] conway.org>

Show quoted text

> PPR happily accepts them; it apparently ignores the /e flag.

I'm close to a solution for that. Show quoted text

> There are some very tricky cases hiding here; in particular, > the RHS of the substitution can contain heredocs and formats and so on.

Indeed. That's why I'm still only "close". ;-) I'll report back as soon as "close" becomes "done". BTW, I doubt PPR will ever successfully parse s/.../.../ee or any greater number of e's, but I suspect that's an acceptable limitation. Damian

Mon Aug 14 07:58:23 2017 The RT System itself - Status changed from 'new' to 'open'

Mon Aug 21 13:44:13 2017 damian [...] conway.org - Correspondence added

Subject:	Re: [rt.cpan.org #122794] s///e RHS is not validated
Date:	Mon, 21 Aug 2017 17:43:22 +0000
To:	bug-PPR [...] rt.cpan.org
From:	Damian Conway <damian [...] conway.org>

The 0.000012 release now handles (most) cases of s/.../.../e correctly, but still has some bad edge-cases. This is an extremely hard problem, which I may not be able to fully solve. I will continue to work on it, whenever inspiration strikes. Damian

Wed Aug 23 01:34:54 2017 $_ = 'spro^^*%*^6ut# [...] &$%*c>#!^!#&!pan.org'; y/a-z.@//cd; print - Correspondence added

On Mon Aug 21 13:44:13 2017, damian@conway.org wrote: Show quoted text

> The 0.000012 release now handles (most) cases of > s/.../.../e correctly, but still has some bad edge-cases. > This is an extremely hard problem, which I may not be > able to fully solve. I will continue to work on it, whenever > inspiration strikes. > > Damian

Good luck! This is something that even Perl cannot get right all the time. Consider this example: s/ /<<END /e END / If you search for the third / you find that there is an /e flag which means that we have a here-doc whose body is " /e\n". But if that /e is part of the here-doc body, then the / at the end must be the one that terminates the s///, in which case, since there is no /e there, we do *not* have a here-doc. I can see why Perl 6 moved the flags.

Wed Aug 23 01:54:06 2017 damian [...] conway.org - Correspondence added

Subject:	Re: [rt.cpan.org #122794] s///e RHS is not validated
Date:	Wed, 23 Aug 2017 05:53:16 +0000
To:	bug-PPR [...] rt.cpan.org
From:	Damian Conway <damian [...] conway.org>

Show quoted text

> Good luck! This is something that even Perl cannot get right all the

time. Indeed. I don't really expect to get this 100% right, especially not in a single pass using regexes. But it's actually heartening to know that it's literally *impossible* to get right. ;-) Show quoted text

> Consider this example:

What a great/terrible example. Thank-you! Show quoted text

> I can see why Perl 6 moved the flags.

We should have been braver...and removed heredocs! They require over 100 lines in the regex just to parse them, and the need to parse them slows the regex down considerably at every newline. :-( Thank-you for the feedback, though. I plan to add your example to my talk on the module. Damian