2006/9/15, François Patte francois.patte@math-info.univ-paris5.fr:

...

People largely expect links to be whole words, not parts of words.

In that case the automatic correction done by wiki is strange: why only one or two syllables? and not to the end of the word?

Is it because, in fact, the "n", in ganita part, is not a latin "n", but a letter with a diacritical mark (underdot)?

Yes. It gets all that has been defined as a letter in the settings. You can change that in [[MediaWiki:Linktrail]] (if you are a sysop). The standard setting is:

/^([a-z]+)(.*)$/sD

If there are other characters than a to z that can occur in words in your language, you can add them before or after the 'a-z'. For example, Dutch uses the setting:

/^([äöüïëéèéàça-z]+)(.*)$/sD

On 9/15/06, François Patte francois.patte@math-info.univ-paris5.fr wrote:

Your example seems (to me) irrelevant: of course nobody wants such a link: [[car]]s; but I explained that it was a compound word and you may want to explain one part of the compound: [[now]]adays (BTW, I don't know what this "a" between now and days stand for!).

Hmm, I can see where this would be annoying in languages with more compound words than English has, and of course it wouldn't be useful in languages with fewer affixes. Thankfully, there's a way for such languages to turn this behavior off in the language files (see next response).

Is it because, in fact, the "n", in ganita part, is not a latin "n", but a letter with a diacritical mark (underdot)? Are wiki's rules fully compliant with utf-8 encoding?

Yes, but yes. A list of characters is maintained per-language to consider "word letters", something which of course is different for each language. Thus, "[[car]]s." will extend to the s but not the period, and "[[lightning]]-rod" won't encompass "-rod". If the language you're using doesn't have that character, please file a bug at http://bugs.wikimedia.org/ stating the language and exactly what characters that language considers letters.

Alternatively, you can suggest that the feature should be disabled for the language (in the same place), but in that case it would probably be a good idea to hold a poll on one of the Wikimedia projects for the appropriate language, or a similar place (this will affect all MediaWiki wikis using this language as their default).

As an aside: why don't we maintain a list of word-characters, namely all Unicode word-character blocks, and allow each language to add to or remove from it as necessary? Would that be too slow? To begin with, currently the behavior is broken on Commons, Meta, etc. for non-English languages; while maybe this can't be avoided entirely (probably some languages will want to include something like ! or what have you), it can be much reduced if the default were something sensible for as many languages as is possible without conflict.

On 9/15/06, Rotem Liss rotemliss_net@fastmail.fm wrote:

Andre Engels wrote:

...

You can change that in [[MediaWiki:Linktrail]] (if you are a sysop).

I think that users can no longer change it in [[MediaWiki:Linktrail]], and they have to set $linktrail in the language file instead.

I believe this is correct.