Hi Clemens, > On 22 Feb 2019, at 22:31, Clemens Robbenhaar <robbenh...@green-meadows.de> > wrote: > > >> Hi devs, >> >> Because of tomcat and because tomcat is by far the most used container for >> xwiki and because we also bundle tomcat in the debian and docker >> dsitributions, I think we should disallow / and \ in page names by default. >> It's causing too much trouble for users. We keep having users posting >> problems and for one user who post a problem there are 100 who don't. >> >> The latest one: >> https://forum.xwiki.org/t/please-help-broken-link-routing-cannot-open-page-from-navigation-any-help-appreciated/4448 > > +1 for me for the stripping of slashes. Even if configured properly there is > a problem with tomcat: if you create a page e.g. in the Sandbox with a title > of "Page A / B", this creates a page with the reference "Sandbox.Page A . > B.WebHome", and you get an "empty" intermediate space.
I’ve just tested with the XWiki Docker image 10.11.3, where Tomcat is correctly configured and it works as it’s supposed to: > > As users mostly see titles in the current XWiki version, it should not > matter to them what happens to the page name. > >> Implementation idea: >> * Offer an extension point in the create page UI to allow plugging some >> cleaning algorithm >> * Provide the \ and / cleaning algorithm by default >> * Handle backward compat. Find ways to not break existing users who are >> having / and \ in page names. For example we could imagine a property in >> xwiki.properties that we would set to point to the hint of the / and \ >> cleaner component. Thus existing users would need to voluntarily upgrade >> their xwiki.properties to have it if they want it. We would need to provide >> some script to convert existing pages with / and \ too probably. >> >> WDYT? > > About backwards compatibility, I do not get the point. What will break for > users who have already existing pages? > As long as the cleaning algorithm only runs on newly created pages there > should not be any bw/compat problem. Yes you’re right. But what’s important is to not change the behavior: if they were using “/“ and “\” in page name we shouldn’t break them and they should continue to do so unless they choose not to. However for new users we can configure them to filter page names by default. Hence my proposal about the new config property in xwiki.properties. When you uprgade you’ll decide if you want to merge it or not and if not you’ll continue to get the same behavior as before. > > If they really want no cleaning, we can provide them with a property which > contains the characters to be scrapped from the name with a default value of > '/\', and if they do not want this, they can set this property to an empty > value. I prefer the implementation that I mentioned based on a component role and provide the hint in the config. It not allows the 2 use cases mentioned above but also any other type of custom use cases (other characters to strip, different logic, like force all creations under a given page, etc). The idea would be to take the reference as input and clean it and generate another reference as output (and display the cleaned version to the user somehow so that there’s no surprise - display the value that’ll be used as a form field hint for ex). > Or what am I missing here? > > Best, > Clemens > >> Thanks >> -Vincent > > P.S. As a side note, when using MySQL creating spaces with a trailing > whitespace seem to cause some odd problems, too. I have seen at least one > occurrence of an entry in the xwikispaces with no corresponding entries in > xwikidoc. This causes an entry in the navigation tree which leads to a > non-existing page and no good way to get rid of it from a users point of view. Yeah just tested and it worked fine for me (the URL is http://localhost:8080/bin/view/Test%20/ ). > I have not figured out how to reproduce the problem, unfortunately. > > (For the issue with trailing spaces try "select 'x' = 'x ', 'x' like 'x '; " > and read https://dev.mysql.com/doc/refman/5.7/en/char.html ) > Vaguely rings a bell. Would be great to open a jira issue if you can reproduce the issue with the query. Thanks -Vincent
