[P5P] [PATCH] New failing test for RT#45667 (m/[#]/x treated inconsistently)

it doesn't seem that op/regexp.t supports TODO tests. That would
probably be a worthwhile addition...

Probably is. But im not entirely comfortable with the original patch.
Its not clear to me that we can change this behaviour anymore. At
least not in the regex engine. Its actually probably easier to change
this in the perl parser, which we long term intend to change anyway in
order to resolve the whole qr// as closure stuff as well as variable
bindings amongst other issues.

Probably a good idea if Dave Mitchell speaks up on this one as to my
knowledge he has the most developed idea of the future of this.

Yves

The bug's definitely in the perl parser, and I still intend to fix it post
5.10.0. The fix is kinda orthogonal to the qr// and closure stuff.
Actually, there's a bit of uncertainty as to whether I cans still do that
qr// stuff, now that there's an API been added to the regex parser: I
don't think it can can handle want I would like to pass to it (a mixture
of const strings and snippets of opcode trees), of if I could pass it,
whether other plugged-in engines (sucha as PCRE) wouldn't choke on it.
Anyway, I haven't looked closely at it yet.

I dont think you should worry about that. I think fixing the general
behaviour takes precedence over supporting experimental regex engine
plug ins. I believe that the API has been documented to be a work in
progress anyway, and if it hasnt we should make sure to do so before
5.10 is released. Of course if what you design has an api that makes
it easy for the engine plug ins to adapt it would be a good thing. :-)

Yves

Well, they're slightly related, but the two fixes can be developed
independently of each other. They both involve making the tokeniser be a
bit more clever when processing literal strings. There's a function that
scans through characters in a literal string until it finds something
"interesting". For example, in the double-quoted string "abc$d", it stops
at the $ and returns "abc" as a token. One fix involves stopping that
function thinking that # is interesting when within [], the other involves
making it think that '(?{' is interesting.
Okay :-)

OK, fair enough. Consistency is what I seek. :-) In that light,
here's an alternative proposed test patch. As of 32199, this patch
yields (abbreviated somewhat):

ok 1342
ok 1343
ok 1344
not ok 1345 () 'foo \[#\]'x:foo[#]:y:$&:foo[#] => `foo[', match=1
not ok 1346 () 'foo [#]':foo#:y:$&:foo# => `'', match=
ok 1347
ok 1348
not ok 1349 () 'foo \[#\]':foo[#]:y:$&:foo[#] => `'', match=

Chris


--- t/op/re_tests.orig 2007-10-28 09:26:00.000000000 -0500
+++ t/op/re_tests 2007-10-30 19:55:16.000000000 -0500
@@ -1338,3 +1338,12 @@
.*\z foo\n y - -
^(?:(\d)x)?\d$ 1 y ${\(defined($1)?1:0)} 0
.*?(?:(\w)|(\w))x abx y $1-$2 b-
+
+'foo [#]'x foo# y $& foo#
+'foo [#'x foo# c - Unmatched [
+'foo \[#]'x foo# n - -
+'foo \[#\]'x foo[#] y $& foo[#]
+'foo [#]' foo# y $& foo#
+'foo [#' foo# c - Unmatched [
+'foo \[#]' foo# n - -
+'foo \[#\]' foo[#] y $& foo[#]

I think there is a way to mark the failing ones as todo tests, youd
have to poke around in the re_tests test file to see how, if not just
modify regexp.t to do it somehow. Probably its just as simple as
saying 'TODO' on the test name, if not, well thats how i would do it.

Im hoping to find some perl related tuits soon.

yves