Created attachment 15697 [details] [review]
patch

I'd rather text/x-graphviz, and the patch is missing a test case, and being added to the tests/list

Created attachment 15737 [details] [review]
same patch + testcase

Same patch with testcase added. I think text/x-dot is better than x-graphviz: graphviz is a collection of tools, but there are also other tools that deal with dot files and the documentaion at http://www.graphviz.org/doc/info/lang.html explicitely refers to the syntax as "the DOT language".

That said, I'll also attach another patch using x-graphviz and let you pick the one you prefer.

Created attachment 15738 [details] [review]
alternative patch

Could you add some magic to the matchlet as well?
From the URL you gave, the file should start with:
"strict graph "
"graph "
"strict digraph "
"digraph "

*** Bug 12677 has been marked as a duplicate of this bug. ***

Please note that .dot files are also msword templates, we need a
mimetype for that one too, with *.dot as a pattern, and magic to differenciate
it from graphviz files (well I guess it could just inherit application/x-ole-storage).

(In reply to comment #7)
> Please note that .dot files are also msword templates, we need a
> mimetype for that one too, with *.dot as a pattern, and magic to differenciate
> it from graphviz files (well I guess it could just inherit
> application/x-ole-storage).

Made it a sub-class of application/msword instead. Added to the test suite.

With your last change, graphviz dot files are now detected as MS files :(

Shouldn't the test suite also check that the dot file I attched is *not* detected as a MS dot file? (and vice versa, that MS test file is not detected as graphiz dot file)

As for the magic pattern for graphviz dot files, I have no clue how it should be done...

is something like this enough?

<match type="string" value="graph" offset="0"/>
<match type="string" value="digraph" offset="0"/>


It really sounds wrong to me since like all source files, most of the time the file starts with comments...

(In reply to comment #9)
> With your last change, graphviz dot files are now detected as MS files :(

Obviously, since I didn't commit your patch!

> Shouldn't the test suite also check that the dot file I attched is *not*
> detected as a MS dot file? (and vice versa, that MS test file is not detected
> as graphiz dot file)

Yes, once we have added your patch.

> As for the magic pattern for graphviz dot files, I have no clue how it should
> be done...
> 
> is something like this enough?
> 
> <match type="string" value="graph" offset="0"/>
> <match type="string" value="digraph" offset="0"/>
> 
> 
> It really sounds wrong to me since like all source files, most of the time the
> file starts with comments...

Then look for it in 0:256, or something like that. And please provide a representative .dot file then.

>> With your last change, graphviz dot files are now detected as MS files :(

>Obviously, since I didn't commit your patch!

mmm... I thought that for binary files *not* matching the magic regex would prevail over a matching glob, otherwise simply renaming junk to .doc etc would make it detected as MS file, no?

(In reply to comment #11)
> >> With your last change, graphviz dot files are now detected as MS files :(
> 
> >Obviously, since I didn't commit your patch!
> 
> mmm... I thought that for binary files *not* matching the magic regex would
> prevail over a matching glob, otherwise simply renaming junk to .doc etc would
> make it detected as MS file, no?

Which is what it does. Read the shared-mime spec for more information about the detection algorithms.

Any news on this?

(In reply to comment #13)
> Any news on this?

I'm still waiting on a patch with decent magic.

In fact this mimetype has been registered to IANA recently (July 2008), under the name text/vnd.graphviz. Good thing we didn't have it in freedesktop.org yet, so we can use the right name for it now ;)

I just committed the definition of text/vnd.graphviz, with globs from IANA (*.gv, *.dot), magic (determined using the dot files I had around), and test (from Paolo Borelli).