In python2, mimetypes module was not deterministic and this snippet:

import mimetypes; print(mimetypes.guess_extension("text/html"))

use to be ".html" on python 2.7.14, but is ".htm" on python 2.7.18

Similarly:

import mimetypes; print(mimetypes.guess_extension("application/msword"))

was ".doc" on 2.7.14 and ".dot" on 2.7.18 (this was in my observations, this does not look deterministic as it is iterating on a dict, maybe this behaviour is not always same)

For html conversion engine, ERP5 expect that the extension for text/html is .html, not .htm. Some tests also expect that the conversion for word is .doc so to keep compatibility with extensions used in ERP5 compatibility mode, define explicitly these two extensions instead of depending on python standard library.