No \G for s///g ?

tye has asked for the wisdom of the Perl Monks concerning the following question:

In Re: Wrap while ignoring certain sequences (from CB) I ended up with this code:

    my $len= 79;
    my $esc= '\e';
    my $eseq= qr[$esc[^a-zA-Z]*[a-zA-Z]];
    my $char= qr[(?:$eseq)*[^$esc\n]];
    my $nonsp= qr[(?:$eseq)*[^$esc\s]];

    s[(?:^|(?<=\s))((?:$char){1,$len}(?:$eseq)*)\s][$1\n]g;
    s[(?:^|(?<=\s))((?:$nonsp){$len}(?:$eseq)*)(?=[^$esc\s])][$1\n]g;
[download]

But it has two bugs (described later). What I think I should use is:

    s[(?:\G|^)((?:$char){1,$len}(?:$eseq)*)\s][$1\n]gm;
    #    ^^^^                                        ^
    #    vv
    s[(?:\G|(?<=\s))((?:$nonsp){$len}(?:$eseq)*)(?=[^$esc\s])][$1\n]g;
[download]

You see, the first substitution should only take place immediately after a newline. The ^ (and m option) take care of starting right after a newline that was already in the string. The \G should take care of starting right after a newline that was just inserted by a substitution (wait for it).

Note that you can't use ^ nor look-behind assertions to detect stuff that was inserted after the s///g started. This isn't documented (that I've seen), but I've tested it (and it makes sense).

\G should say "start where I left off last time" -- except double checking the documentation, it appears that this is only supported for m//g, not s///g. Is there a good reason for this that I'm missing? Update: Appears to be just a documentation issue (perhaps even an issue just with my reading of docs). See my reply below.

So the first substitution in the first block of code above could be better because it wastes time trying to split each short line starting at each whitespace in the line. The (?<=\s) works by matching the space that we just turned into a newline.

The second substitution in the first block of code is broken because a single word that spans more than two lines will only be wrapped once. I want to allow substitutions that start either right after I've wrapped a long word, or at the start of the next word. But without \G, I don't see any way to allow starting right after a substitution that doesn't also allow starting anywhere (which creates another bug as discussed on my previous node).

Anyone see a way I can work around this missing feature? Should I report this as a "bug"?

BTW, playing with this code is made easier by setting $len much lower and making $esc a regular character.

- tye

Comment on No \G for s///g ? Select or Download Code

Replies are listed 'Best First'.
Re: No \G for s///g ? by blakem (Monsignor) on Mar 21, 2003 at 21:13 UTC
\G ... is only supported for m//g, not s///g I didn't dig into your regex, but I think \G is supported for s///g; `$ perl -le '$_="abc def"; print; s/\G\w/X/g; print' abc def XXX def` [download] -Blake	[reply] [d/l]
Re^2: No \G for s///g ? (oops) by tye (Sage) on Mar 21, 2003 at 21:52 UTC
Thanks! I thought I'd used that before but perlop having several examples of m/\G.../g and no examples of s/\G...//g and no mention of \G in the s/// section combined to make me think maybe that was the problem (as I was posting -- writing up a problem always brings new solutions to mind, ya know). I verified that this works on my version of Perl as well. I'd been meaning to post this for a week and finally had a few minutes while waiting for stuff to compile. This response got me to take some time I didn't have to write up some test cases. I must have had an unseen bug in the code when I was testing before because it works fine now. Thanks, blakem. - tye Read more... (4 kB)	[reply] [d/l] [select]
Re: Re^2: No \G for s///g ? (oops) by demerphq (Chancellor) on Mar 22, 2003 at 12:52 UTC
Try setting $len to 5. You'll see that the 'tight' version fails to wrap correctly under certain circumstances like this `Wrapping: @[0;7mCoruscate@[0m says this is a test of the line wrapping code Tight code1: @[0;7mCoruscate@[0m says this is a test of the line wrapping code Tight code: @[0;7mCorus cate@[0m says this is a test of the line wrapp ing code` [download] which suggested to me that the regexes be executed in opposite order `Wrapping: @[0;7mCoruscate@[0m says this is a test of the line wrapping code Tight code1: @[0;7mCorus cate@[0m says this is a test of the line wrapp ing code Tight code: @[0;7mCorus cate@[0m says this is a test of the line wrapp ing code` [download] which seems to do the trick. I believe the first regex gets stymied by lines that have to be split many times at a space, as well as being split many times inside a word. --- demerphq	[reply] [d/l] [select]
Re^4: No \G for s///g ? (oops) by tye (Sage) on Mar 22, 2003 at 19:59 UTC


Syntactic Confectionery Delight
	PerlMonks