At first I’d need some reliable source of such magic numbers and their mapping to MIME-types.
it’s already on your computer
/usr/share/mime/magic contains the compiled informations about magic numbers while /usr/share/mime/globs all the glob patterns.
file --mime-type results are as reliable as the databases on your system is good.
The database is populated by the applications installing their own mime specifications or manually by creating the correct XML files and then running update-mime-database.
but I don’t think that a sequential pattern match is fast enough.
On my system magic is not that big
wc -l /usr/share/mime/magic
1767 /usr/share/mime/magic
Half of the lines are taken by mime definitions, so in total it’s less than a thousand patterns.
wc -l /etc/mime.types
846 /etc/mime.types
Even testing all of them sequentally should be fast enough.
