Wouldn't it make more sense to create a global mapping for all users based on some keywords in the message? I.e. the source code should be improved. Interested in providing a pull request?

Or is your intended categorization not usable for other users?

I'm not sure if everyone would have the same prio view as we have.

The warnings emitted are like:

   "<srcfile>", line <nr>: warning #179-D: variable "some_var" was declared ...
   "<srcfile>", line <nr>: warning #64-D: shift count is too large

Which I personally would classify as different severities.
It would be nice if it were poossible to supply maybe a 'classification overrules' file per compiler containing a means to overruled the default assigned severity by means of a regex ?

In that way anyone can use the mechanism for any compiler, while keeping a relative simple implementation.

I'm not sure if I would be the best person to implement a patch (esp if it is Java code); I've never coded anything in a plugin... which code precisely are you referring to that could be improved ?

The code is a link

If we make this not part of the distribution: what would be a good way to provide such a mapping? A configuration file, a UI, etc.?

my preference would be a (optional) config file. Not sure if there is some kind of standard for config files (XML,INI, ...)?

The config file would contain a set of rules with afaict a regexp and the classification that should give. I would opt for implementation where the first rule-match defines the result. Something like a if/elif list construct. And no match then gives the currently implemented default.