Bad string replacement of entities in build result

I just noticed that the output of maven is wrongly encoded. Here's one of lines of he generated html source :
[INFO] Setting property: resource.loader => 'classpath'.

Remind: In the console, the output is the following (output of mvn site) :
[INFO] Setting property: resource.loader => 'classpath'.

As you can see, there's a problem
I guess the entity conversion is done twice :

• first "=>" is converted "=>"

then another pass detects the & although it shouldn't process it anymore, so :

• second "=>" is in turn converted "=>"

I guess the correction weight might depend on how those conversion are done. As a lot of conversions produces an &, maybe encoding & as & should be the very first conversion to be done in the list ?
Maybe even some "dependency" tree could do the trick (or even better, an API that already does that neatly ).

Cheers.

Let me know if you need anything else.