Ian Hickson wrote:
I would strongly recommend that specs disallow sniffing of any kind, treating all metadata as authorative (and giving explicit priorities to handle metadata contradictions).
I guess the issue is what do you consider metadata? I consider the fact that a file starts with the magic number for a gzip stream a very reliable piece of metadata (much more reliable than HTTP headers - oh and much more reliable than searching a document in an unknown encoding for 'html'). So, in my view you would have a metadata contradiction.
In certain rare cases where sniffing is required due to missing metadata (as in the lack of byte order information when no authorative default is stated), then the sniffing performed should be well defined in order to allow all UAs to have exact interoperability.
I would actually be _very_ happy to see the SVG spec define how this should be done, and in fact I think the portion of the spec originally quoted was doing this. The problem is that for some bizzare reason people don't seem to want to consider the file it's self a good source of metadata about what the file contains. For freeform text files I can understand this but gzip is a binary format it's structure is very rigid and if it is not a gzip stream you will find out quite quickly, not to mention it even includes a CRC of the uncompressed data, it is essentially beyond reason to consider a case where a gzip stream uncompresses successfully and it is not in fact a gzip stream. Now if what you are left with is an SVG document or not is a much tricker proposition.