[pmwiki-users] [pmwiki-devel] RFC: Community poll for PageList link/category arguments

[Petko Yotov]

On 27/12/2021 07:57, Petko Yotov wrote:
> When you're searching for links to multiple pages or to wildcards, the
> new core algorithm is different from the one in the recipe, with fewer
> loops, and I believe (hope) it will be faster than the current recipe.

On PmWiki.org we only have about 10K pages, and here is our benchmark:

   https://www.pmwiki.org/wiki/Test/CorePLMT2

Single page:
   link=PmWiki.BasicEditing

17: 00.22 00.21 PageListTermsTargets begin count=10109
20: 00.25 00.24 PageListTermsTargets end count=70

That's 0.03 seconds.


Wildcard, multiple pages:
   link=PmWiki.BasicEditin?,NoSuchGroup654.NoSuchPage987

52: 00.32 00.30 PageListTermsTargets begin count=10109
55: 00.39 00.37 PageListTermsTargets end count=72

That's 0.07 seconds, only 0.04s or 1/25th of a second slower than the 
first one. I think it might be good enough for many people.

Note that the second search is case-insensitive, it returns 2 additional 
pages which link to Pmwiki.BasicEditing (lowercase "w" in "Pmwiki") 
which doesn't exist.

Not sure if I should do something about this case, or not.

Petko

On 27/12/2021 07:57, Petko Yotov wrote:
> Thank you Gilles, these are great questions. My replies below.
> 
> On 27/12/2021 01:28, ABClf wrote:
>> I understand current installations won't be broken, yet I have a
>> question about performance, concerning multiple link in pagelist :
>> when benchmarking the same pagelist, targeting a one single link,
>> native link=mylink against cookbooked links=mylink,
>> it looks to me that native is faster ; am I right ?
>> 
>> https://www.languefrancaise.net/Test/PageListMultiple (around 3)
>> https://www.languefrancaise.net/Test/PageListNormal (around 1,70)
>> 
>> If so, would it be possible to get the best of each : native for one
>> target search, and complex for complex queries ?
> 
> First, this is a fabulous installation with 64817 pages! It will be
> great indeed to benchmark the new feature.
> 
> Second, we need to look at the numbers between the start and end of
> the functions, so the difference between the core and the recipe are
> even greater:
> 
> 422: 00,69 PageListTermsTargets begin count=64817
> 425: 01,09 PageListTermsTargets end count=164
> 
> 1.09-0.69=0.4 seconds for the core processing
> 
> 422: 00,76 PageListMultiTargets begin count=64817
> 425: 03,17 PageListMultiTargets end count=164
> 
> 3.17-0.76=2.41 seconds for the recipe, that's about 6 times more 
> CPU-intensive.
> 
> 
> Indeed, in 2.3.0, when you have link=SinglePage, then the original
> scanning method is used, so the times should be very close to the
> current ones.
> 
> When you're searching for links to multiple pages or to wildcards, the
> new core algorithm is different from the one in the recipe, with fewer
> loops, and I believe (hope) it will be faster than the current recipe.
> 
> Again, your installation will be a great place to verify this.
> 
> 
>> Question 2, for this paragraph :
>> 
>> link=PageA,PageB should list pages linking to "AT LEAST ONE" among
>> PageA and PageB.
>> link=+PageA,+PageB should list pages linking to "BOTH" PageA and 
>> PageB.
>> 
>> I would like to know if case bellow is involved in the choice.
>> In my case, I have a group made of pages named 1900, 1901, 1902, etc.,
>> for aggregating words by years.
>> Then, for the query
>> links=Year.199*
>> expected output would be to list pages having a link to any year page
>> from 1990 to 1999 (and not list pages having a link to all pages
>> included in the range).
> 
> link=Year.199* or link=+Year.199* lists pages linking to any page with
> the pattern, not to all pages, whether the target page exists or not.
> 
> The wildcard characters are about characters, not about existing pages:
> 
>   ? = exactly one character, any character
>   * = between zero and any number of any character(s)
> 
> 
> The poll is more about a case like link=GroupA.*,GroupB.* then we need
> to decide how we want it to behave.
> 
> 
> Should it require at least one link to GroupA OR to GroupB like in the 
> recipe?
> 
> Should it require at least one link to GroupA AND at least one link to
> GroupB? Or should this case be written as link=+GroupA.*,+GroupB.* ?
> 
> 
> Petko
> 
>> Le sam. 25 déc. 2021 à 09:13, Petko Yotov <5ko at 5ko.fr> a écrit :
>>> 
>>> Hello,
>>> 
>>> I hope you are all safe, and happy holidays if you're celebrating!
>>> 
>>> I've been working on some new features for version 2.3.0 and I'd like
>>> your input.
>>> 
>>> If you have a few minutes, please look at this page:
>>> 
>>>    https://www.pmwiki.org/wiki/PITS/01475
>>> 
>>> ...and consider which ones of the proposed implementations may suit 
>>> your
>>> wiki and your editors best, and add your name.
>>> 
>>> I'd like to see some comments/votes and to release 2.3.0 by January 
>>> 8.
>>> 2022, which will include implementing and documenting the new 
>>> features.
>>> 
>>> If you have questions or suggestions, you can add them to the page, 
>>> or
>>> reply to the mailing list.
>>> 
>>> Thank you in advance.
>>> 
>>> Petko
> 
> _______________________________________________
> pmwiki-devel mailing list
> pmwiki-devel at pmichaud.com
> http://www.pmichaud.com/mailman/listinfo/pmwiki-devel