Description: ------------ One month ago, I've give attention (on PEAR QA mailing list) on a strange behaviour about PEAR_Info end-user doc link, and Helgi gave me an answer. [1] Sadly, this event come again. Perharps it's time (officially) to see what was wrong in cron job, and really fix the situation. It's sad for me to have wrote a doc / include it in PEAR manual, and realized that some users do not read it because a link was not fixed. Regards Laurent [1] http://news.php.net/php.pear.qa/4755

I believe this is due to an over ambitious find-documentation.php doing an UPDATE query without checking what is already stored in the database. My initial tinkering around this quickly fizzled... :S

I believe the update is needed ... lets say pear_info would change their place in the manual ... would be just as bad not to update the url as removing it like it currently does, what we have to do is to figure out why this happens, I might need to checkout peardoc and give it a whirl. More checking is required in the script, no doubt.

The more I'm looking at this the more I'm tempted to rejig the storage / api for this. Goal: Keep package administration simple, and discover links automatically for pear packages. Discover: 1) Iterate over all known packages 2) Look in the relevant documentation path (/docs/package_name/ 3) Foreach xml_files as file { $links = Documentation::discoverLinks($text); //inspects text from somewhere foreach ($links as $link) { Package::addDocumentationLink($package_name, $link); } } Administrate: 1) Provide a Package::getDocumentationLinks($package_name); which returns all available links - nothing is deleted automatically 2) Provide Very Simple UI to actually delete a link from the user interface for package administrators That would prevent the current problems of: - External documentation can get deleted - The current cron is based on the available documentation files, not the available packages - We're actually retrieving the external link and deleting if its 404ing; but Missing Isn't Broken - The current code is very difficult to unit test

The attached patches are a first draft of a new way of finding documentation. Remaining tasks - Do we reimplement the fetch-url-and-check-it-exists behaviour? - Solid testing - Use of the appropriate database objects - Think of a better way to set the 'primary' documentation link - Consider new UI for maintaining this new data

With new host euk1 and new package doc page, this bug should be close without doing nothing else

Helgi sez: Looks like the move to a new server auto fixed this.