On Fri Jul 03 12:29:34 2015, masak wrote: > <hoelzro> I found some behavior using heredocs, and I'm not really > sure what the right thing to do is > <hoelzro> https://gist.github.com/hoelzro/3b1ff9951908c9ce5aa4 > > (Gist shows the following heredoc with a \n in it, which gives a > de-indentation warning at compile time.) > > my $here = qq:to/END_TEXT/; > foo\nbar > END_TEXT > > <hoelzro> it's concerning how leading whitespace removal is handled > when one has an embedded newline via \n in a heredoc > <timotimo> oh, interesting > <timotimo> i hadn't considered newlines that are escaped inside > heredocs like that when i wrote the new dedenting code > <hoelzro> timotimo: I hadn't considered it either until I was > converting code that used a single non-heredoc line and came across > that =/ > <hoelzro> I would file a ticket, but I don't even know if that's a > bug. I don't even know what I expect it should do! > <lucasb> Is there a way to dedent the heredoc literal text before > expanding newline escape sequences? > <hoelzro> TimToady may have to weigh in > <masak> I'm... not sure I feel that's a bug... > <hoelzro> masak: "not sure" is the thing with me too =/ > <masak> I can understand the argument that this \n newline is not part > of the heredoc's actual newlines, but... > <masak> ...I'm also not ready to tear up compiler internals just to > fix this one. > <masak> but let's see if TimToady has some nice insight about it. > <masak> it's kind of related to (but not the same as) > http://strangelyconsistent.org/blog/here-be-heredocs under the heading > "Indentation, revisited" > <colomon> why would you want to embed a \n in a heredoc? > <masak> colomon: that's a good, but separate question :) > <masak> we're exploring "what happens when you do?" > <masak> I don't believe it should be illegal. > <hoelzro> colomon, masak: there's no good reason, I just found it > interesting > <masak> it is interesting. > <hoelzro> I was converting a long single double quoted string with > multiple embedded newlines into a heredoc while refactoring a program > <masak> I mean, from the user's point of view, a \n in a heredoc isn't > the same as a hard newline in a heredoc. > <hoelzro> right > <colomon> what happens if you embed a string with a newline in it? I > actually do that quite a bit in my (mostly p5) scripts > <masak> colomon: read the blog post I linked. > <jnthn> I seem to remember I did it one way in the original herdocs > impl in Rakudo, and something was off and got changed (probably what > masak++ is referencing) > <masak> yes. > <TimToady> re \n in heredocs, the dedent should be happening before \n > interpolation, just as it happens before $I-contain-a-newline > interpolation > <TimToady> so re http://irclog.perlgeek.de/perl6/2015-07-03#i_10845432 > I'd call it a bug > > (URL referencing hoelzro's original example at the top of this report) > > <TimToady> recall that dedent is only for the purposes of establishing > a false margin on the literal text, so from the user's perspective it > really needs to happen before any kind of string processing > <TimToady> I can think of several ways to fix it > <TimToady> one is to treat \n interpolation more like $newline > interpolation > <TimToady> another is to substitute a sentinel character for \n inside > heredoc parsing, and only dedent after the sentinel (and replace the > sentinel with a real \n) > * masak submits TimToady-approved heredoc \n de-indentation rakudobug
Turns out there was a neater way than a sentinel: in the case of being inside a heredoc, code-gen \n and \r\n as an AST node, not just a string. This triggers the same codepaths that deal with interpolation. Tests in S02-literals/quoting.t. /jnthn
