How do I assign my regexp-result to $& ?
$& is read-only
irb(main):001:0> $& = “test”
it depend what you want to do
/test/ =~ “test” # $& ==> “test”
···
–
Guy Decoux
Any ideas how to fake this?
Well, it’s easy to do for $’ and $`, no idea for the rest $1, …
Any rationale why they are write-protected?
These variables are the result of a regexp against a string : it seems a
non sense to change this result.
···
–
Guy Decoux
“Simon Strandgaard” wrote:
That will work. But I also need to so assignment to $1 - $9, $', $`, $+
class MatchData
def MatchData.(pre, match, post, *captures)
str = [pre, match, post] * ‘’
match = Regexp.quote(match.dup)
captures.collect! do |cap|
c = Regexp.quote(cap)
i = match.index(c) or raise ArgumentError,
"Not a backreference: #{cap.inspect}"
match[i + c.size, 0] = ')'
match.slice!(0, i) + '('
end
str =~ /#{captures * '' + match}(?=.{#{post.size}}\z)/m
$~
end
end
$~ = MatchData[*%w/alpha beta charlie beta be e ta t a/]
p [$`, $&, $', *$~.captures]
Aree, these values should only contain output from regexp.
But as a paradox I am developing my own regexp engine!
I want my regexp engine to be compatible with Ruby’s, therefore I want to
exercise the engine with the Rubicon testsuite. But rubicon uses $& and
$1-$9. I can substitute all occurencies of $& and $1-$9 in rubicon.
If $& and $1-$9 assignment are impossible, then, then my regexp-engine
never will get compatible with Ruby’s engine 
[RCR] remove write-protection of $&, $1-$9, $', $`, $+
or perhaps let them stay write-protected, but let assignment via const_set
become possible?
···
On Wed, 19 Nov 2003 13:49:27 +0100, ts wrote:
Any ideas how to fake this?
Well, it’s easy to do for $’ and $`, no idea for the rest $1, …
Any rationale why they are write-protected?
These variables are the result of a regexp against a string : it seems a
non sense to change this result.
–
Simon Strandgaard
“Simon Strandgaard” wrote:
That will work. But I also need to so assignment to $1 - $9, $', $`, $+
[snip fake MatchData class]
$~ = MatchData[*%w/alpha beta charlie beta be e ta t a/]
p [$`, $&, $', *$~.captures]
Weew, nice. I were starting to believe that this were simply impossible.
Surprise of today. I am grateful for your help. Thanks.
···
On Sun, 23 Nov 2003 06:20:06 -0800, Sabby and Tabby wrote:
–
Simon Strandgaard
I want my regexp engine to be compatible with Ruby's, therefore I want to
exercise the engine with the Rubicon testsuite. But rubicon uses $& and
$1-$9. I can substitute all occurencies of $& and $1-$9 in rubicon.
Be carefull with this test : it was designed for the *current* regexp
engine, this means, for example, that these tests must be modified for
Onigurama. Because Oniguruma will give different results.
[RCR] remove write-protection of $&, $1-$9, $', $`, $+
No need for this : just write an C extension which do this, and use this
extension *only* for testing your regexp engine.
Nobody, except you, need to modify these variables.
Guy Decoux
Hi,
···
In message “Re: $& write-protected?” on 03/11/19, Simon Strandgaard qj5nd7l02@sneakemail.com writes:
[RCR] remove write-protection of $&, $1-$9, $', $`, $+
or perhaps let them stay write-protected, but let assignment via const_set
become possible?
If you update $~ (the match result), $&, $1, etc. would follow
accordingly. So that the better RCR should be to create MatchData
without built-in regular expression match, I think.
matz.
I want my regexp engine to be compatible with Ruby’s, therefore I want to
exercise the engine with the Rubicon testsuite. But rubicon uses $& and
$1-$9. I can substitute all occurencies of $& and $1-$9 in rubicon.Be carefull with this test : it was designed for the current regexp
engine, this means, for example, that these tests must be modified for
Onigurama. Because Oniguruma will give different results.
I have about 150 testcases which I both test with Ruby’s current regexp
engine and against my own engine. So far my engine is fully compatible.
Perhaps I should also run the tests agains Oniguruma.
Do you know if Oniguruma can coexist with Ruby’s current regexp engine?
[RCR] remove write-protection of $&, $1-$9, $', $`, $+
No need for this : just write an C extension which do this, and use this
extension only for testing your regexp engine.Nobody, except you, need to modify these variables.
Great, so it is possible… I will look at it tomorrow (C++SWIG).
···
On Wed, 19 Nov 2003 23:01:50 +0900, ts wrote:
–
Simon Strandgaard
Good idea for an RCR. I vote yes immediately for a pure MatchData class 
···
On Wed, 19 Nov 2003 23:18:12 +0900, Yukihiro Matsumoto wrote:
In message “Re: $& write-protected?” > on 03/11/19, Simon Strandgaard qj5nd7l02@sneakemail.com writes:
[RCR] remove write-protection of $&, $1-$9, $', $`, $+
or perhaps let them stay write-protected, but let assignment via const_set
become possible?If you update $~ (the match result), $&, $1, etc. would follow
accordingly. So that the better RCR should be to create MatchData
without built-in regular expression match, I think.
–
Simon Strandgaard
If you update $~ (the match result), $&, $1, etc. would follow
accordingly. So that the better RCR should be to create MatchData
without built-in regular expression match, I think.
Here, here!
I was trying to add some features to the regex engine and couldn’t for
similar reasons. (I was trying to leverage the built-in syntax for
regexes, but to match against tree-shaped structures… Don’t ask.)
I eventually gave up because I couldn’t create MatchData.
Ari
Perhaps I should also run the tests agains Oniguruma.
yes, like said previously, Oniguruma will give you different results.
Do you know if Oniguruma can coexist with Ruby's current regexp engine?
well, you can have 2 differents ruby : one with Oniguruma, the other with
the GNU regexp
Great, so it *is* possible.. I will look at it tomorrow (C++SWIG).
^^^^
no need of SWIG in this case 
Guy Decoux
Great, so it is possible… I will look at it tomorrow (C++SWIG).
^^^^
no need of SWIG in this case
Maybe overkill.
My first thought were to extern ‘last_match_getter’ from ‘re.c’ and
just supply my own last_match_setter. But then I noticed the function are
are ‘static’ (private).
How do you think I shall implement this ?
gcc assign.c -I/usr/home/neoneye/install/ruby-1.8.1
/usr/lib/crt1.o: In function_start': /usr/lib/crt1.o(.text+0x81): undefined reference tomain’
/var/tmp//ccu9UXrr.o: In functionInit_Assign': /var/tmp//ccu9UXrr.o(.text+0x17): undefined reference tolast_match_getter’
/var/tmp//ccu9UXrr.o(.text+0x21): undefined reference to `rb_define_virtual_variable’
expand -t4 assign.c
#include <ruby.h>
extern VALUE last_match_getter();
static void last_match_setter(VALUE val) {
/* TODO */
}
void Init_Assign() {
rb_define_virtual_variable(“$&”, last_match_getter, last_match_setter);
}
···
On Wed, 19 Nov 2003 23:24:17 +0900, ts wrote:
My first thought were to extern 'last_match_getter' from 're.c' and
just supply my own last_match_setter. But then I noticed the function are
are 'static' (private).
see the struct RMatch in re.h. str is the string, regs are the
registers. ruby use register to retrieve $&, $1, ...
see for example rb_reg_match_pre() ($'), rb_reg_match_post() ($`),
rb_reg_nth_match() ($&, $1, ...)
If you change the struct RMatch, this change will be reflected on $&,
$1, ...
Guy Decoux
I bail out, maybe I will look at it tomorrow. I hoped it wouldn’t be this
complicated (im tired). I will follow the other approach and substitution
all occurencies of $& and $1-$9 instead.
Thanks for sharing you wisdom.
BTW: Is there a reason that functions not found in Ruby’s .H files, often
has a ‘static’ tag, so thats impossible to extern them?
Would it make any sense to remove all ‘static’ tags?
···
On Thu, 20 Nov 2003 00:25:59 +0900, ts wrote:
My first thought were to extern ‘last_match_getter’ from ‘re.c’ and
just supply my own last_match_setter. But then I noticed the function are
are ‘static’ (private).see the struct RMatch in re.h. str is the string, regs are the
registers. ruby use register to retrieve $&, $1, …see for example rb_reg_match_pre() ($'), rb_reg_match_post() ($`),
rb_reg_nth_match() ($&, $1, …)If you change the struct RMatch, this change will be reflected on $&,
$1, …
–
Simon Strandgaard