Front page | perl.perl5.porters |
Postings from November 2010
On 1 November 2010 00:18, Father Chrysostomos via RT
<perlbug-comment@perl.org> wrote:
> On Thu Oct 28 09:50:54 2010, demerphq wrote:
>>
>> /me goes back to sleep
>>
>> yves
>
> Wait! could you look at bugs 68564, 70998 and 78356 first? Or am I too
> late? :-)
68564: Your patch doesn't *feel* right to me, like its addressing the
symptom and not the cause. I need to poke deeper later.
70998: Similar story. As far as i remember
/\xab|\xa9/
should be handled by the trie optimisation, and then converted into a
charclass (ANYOF). So the question is why doesnt this problem also
occur with
/[\xab\xa9]/
? Im not saying the patch is wrong, just it doesnt clarify this aspect
of things.
78356: Ihis is an interesting case. IMO there are a few things at play
here. First is that we go through a phase during optimisation where we
join consequtive EXACT nodes together. The original code used to
produce one EXACT node per character, later it was realized that
merging them made a big difference. However it seems the case of
NOTHING EXACT doesnt get reduced (properly) to EXACT alone.
IIRC that is what this output is:
| 13| tail~ BRANCH (9) -> TAIL
| | tsdy~ NOTHING (3) -> PSEUDO
| | ~ EXACT <foo> (4) -> EXACT
| | ~ attach to TAIL (12) offset to 8
| | tsdy~ EXACT <bar> (7) -> EXACT
| | ~ attach to TAIL (12) offset to 5
| | tsdy~ EXACT <zot> (10) -> EXACT
| | ~ attach to TAIL (12) offset to 2
Next, it appears that the jump trie code isnt working for the case of
NOTHING. which is a separate problem.
We see this in action here:
Final program:
1: SBOL (2)
2: TRIE-EXACT<S:1/7 W:3 L:0/3 C:6/6>[bz] (13)
<> (4)
4: EXACT <foo> (13)
<bar> (13)
<zot> (13)
13: EOS (14)
14: END (0)
anchored ""$ at 3 anchored(SBOL) minlen 3
r->extflags: ANCH_SBOL
Matching REx "\A(?:(?:)foo|bar|zot)\z" against "foo"
Setting an EVAL scope, savestack=3
regmatch start
0 <> <foo> | 1:SBOL(2)
0 <> <foo> | 2:TRIE-EXACT<S:1/7 W:3 L:0/3 C:6/6>[bz](13)
matched empty string...
0 <> <foo> | 13:EOS(14)
failed...
Where we expect to execute node 4 after the "matched empty string",
yet we did not. For some reason we didnt "jump" but execute a "normal"
trie.
Ill try to dig further later on when I have more time. I hope this is
helpful to you.
cheers,
Yves
--
perl -Mre=debug -e "/just|another|perl|hacker/"
Thread Previous