Front page | perl.perl5.porters |
Postings from April 2001
I'm afraid I've not kept up to date with the discussions on this, but
the latest approach (@9846) isn't working: it changes regcomp.c:S_dumpuntil
to use 'node = next', which relies on node->next_off, which isn't set.
This is giving me coredumps:
crypt% ./perl -Dr -we 'warn "match <$&>\n" while "xbx" =~ m{(\G|x)([^x]*)x}g' 2>&1 | more
Compiling REx `(\G|x)([^x]*)x'
size 28 Got 228 bytes for offset annotations.
first at 3
rarest char x at 0
1: OPEN1(3)
3: BRANCH(5)
4: GPOS(8)
5: BRANCH(8)
6: EXACT <x>(8)
8: CLOSE1(10)
10: OPEN2(12)
12: STAR(24)
13: ANYOF[\0-wy-\377{all-unicode}](0)
Segmentation fault (core dumped)
crypt%
I think that the right solution is to ensure node->next_off is set
correctly for ANYOF nodes, as in the patch below. This fixes the
coredump, so the above test loops as expected (working on that one).
Hugo
--- regcomp.c.old Sun Apr 22 17:17:39 2001
+++ regcomp.c Thu Apr 26 05:28:56 2001
@@ -3929,6 +3929,7 @@
n = add_data(pRExC_state, 1, "s");
RExC_rx->data->data[n] = (void*)rv;
ARG_SET(ret, n);
+ NEXT_OFF(ret) = RExC_emit - ret;
}
return ret;
Thread Next