[pmwiki-users] How good (and usefull) is AI at writing basic (or not so basic) PmWiki code ?

[Petko Yotov]

On 13/05/2025 14:56, ABClf wrote:
> Markup('ExtractLine', 'directives',
>   '/\\(:ExtractLine\\s+(\\S+)\\s+([\'"])(.+?)\\2\\s*:\\)/',
>   'ExtractLineMarkup');
> 
> function ExtractLineMarkup($m) {
>     $targetPage = $m[1];
>     $pattern = $m[3]; // Motif brut sans délimiteurs
>     return ExtractAllMatchingLines($targetPage, $pattern);
> }
> 
> function ExtractAllMatchingLines($pagename, $pattern) {
>     $page = ReadPage($pagename, READPAGE_CURRENT);
>     if (!$page) return "Page non trouvée";
> 
>     $text = $page['text'];
>     $lines = explode("\n", $text);
>     $matches = [];
> 
>     foreach ($lines as $line) {
>         if (@preg_match("/$pattern/", $line)) {
>             $matches[] = Keep($line);  // Empêche le parsing wiki
>         }
>     }
> 
>     if (count($matches) > 0) {
>         return implode("<br />\n", $matches);
>     } else {
>         return "Aucune ligne ne correspond au motif.";
>     }
> }
> 
> Might be easy for smart people, but I'm still so admirative. For the
> result I get so quickly, and for the easy way to get it, fluently, in
> 3, 4 "conversational" steps. That's not something I would have
> expected 1 year ago. One more second and I believe you get the
> extracted line cleaned up from PmWiki markup if you need it.
> 
> My question is two open questions :
> 
> 1. is the Ai generated code well done ? Ok, it works, but is it well
> written ? secure enough ? doesn't it forget something important ?
> efficient enough for speed ? (I believe it is, because here it is
> something very basic).

This is indeed impressive. The only major problem I see is that it 
doesn't check for permissions on the page, it should use

   RetrieveAuthPage($pagename, 'read', false, READPAGE_CURRENT);

instead of ReadPage().

Currently if different people have different permissions, someone typing 
in a sandbox and pressing Preview could retrieve all text from SiteAdmin 
like emails of the other users in Notify, possibly sensible AuthUser 
information, or any text from any other protected page. And you'll never 
know.

Less dangerous but may be annoying, you need to type the full Group.Page 
in the markup, not relative names like Page (in the same group).

The speed is not really a problem, but using PmWiki helper functions 
would make the code shorter, more readable, and easier to maintain and 
update.

A minor improvement for easier maintenance and evolution might be to use 
ParseArgs() to parse the arguments. You don't need to quote arguments 
that do not contain spaces, "=" or ":".

Also there is $MarkupDirectiveFunctions that makes it even 
easier/simpler to add a directive, receive its parsed arguments, and 
update it in the future with new arguments.

Also for convenience, it may be useful to allow simpler "glob" patterns 
like ? and *. The MatchNames() function will match all lines at once 
against a regular expression if the pattern starts and ends with "/" or 
"!" (inverted match), otherwise a glob pattern will be assumed:

   "/^(Hello|Bonjour)/" -> find lines starting with Hello or Bonjour

   "Hello*,Bonjour*" -> find lines starting with Hello or Bonjour
   "Hello*,-*Bye*" -> starting with Hello that do not contain Bye

BTW there is Cookbook:Grep.

> 2. do you play with AI tools like ChapGPT, Deepseek, Mistral for
> improving your PmWiki ? What do you do (can we do something else than
> markups and recipes ?) Does it work fine for you ? Which AI do you
> find is the best for this kind of work ?

I have tried ChatGPT but found that for code it takes me more time to 
explain, review and correct its output, than if I wrote it myself. Never 
for writing PmWiki code, I'm surprised it was able to assemble something 
that actually works.

ChatGPT works for my linguistic questions, like explain to me an English 
language expression I don't understand or a cultural reference I don't 
get.

Petko