Hello,


[original scenario]
> 1. Let doc.html be an HTML 4.01 document served as "text/html".
> 2. The content of doc.html is encoded with UTF-8, and
>    the meta element declaration correctly says so.
> 3. The document also contains a "script" element having
>    type="text/javascript" and src="ext.js", but no
>    "charset" attribute.
> 4. The HTTP server does not serve a "charset" header,
>    neither for "doc.html" nor for "ext.js".


Thanks to James Justin?s answer, I see a possible explanation now.

I?ll try to sum it up myself:

- Authoring a script ref. to a non-US-ASCII external resource REQUIRES
   the author to specify a charset (as RFC 2046 demands). This can be
   done either way: embedded in the "type" attribute or by specifying
   a separate "charset" attribute within the script element.

- In particular, declaring a character encoding *for the
   entire document* is not sufficient to fulfill the above requirement.
   User agents are not REQUIRED to propagate, for example, a charset
   declaration given in the META element all the way down to the content
   of the external script resource.

- However, the user agent MAY take the document encoding (e. g. from the
   META element) as a fallback/hint. It seems to me that most
   (but not all) popular browsers do exactly that.
   (This is what actually raised my question in the first place.)

- Lesson learned: Given the scenario, the user agent behaviour
   is not specified according to HTML 4.01.

I hope this explanation is roughly correct.


Regards
Claudio Pellegrino