> On Wed, Mar 1, 2017 at 5:35 PM, Dagfinn Ilmari Mannsåker <ilmari@ilmari.org> > wrote: […] >> #2 MIME-decodes headers received from the mailing list archive JSON API >> >> I haven't been able to talk to the JSON api, so I couldn't test them >> properly, but I did some stand-alone testing of the code snippets. >> >> Note that the MIME decoding only works properly if running under Python >> 3; the Python 2 version of email.header.decode_header() has broken >> detection of the end of encoded-words. > > Is the patch still an improvement on python2?
No, because it'd be affected by the same problem that causes the undecoded headers to be returned from the archive app.
OK.
> Also, based on your other email about the list archives -- if we fix this > in the archives, does that make this patch unnecessary?
Yes, this patch is unnecessary if the archive app is fixed, and insufficient if the commitfest app isn't upgraded to python3.
One possible workaround until upgrading to python3 is feasible would be for the archive app to do some more munging (akin to the existing _re_mailworkaround), and inject a space between an encoded-word and an immediately-adjacent opening/closing paren.
Actually, if I read that one right, it would be enough to upgrade the *loader* part of the archives, which is a much more contained problem, as it pretty much only has dependencies on the standard library.
Will have to run some detailed tests on that of course, to make sure it doesn't break anything else (like we have to reparse the 1.2 million messages in the archives and see if something else changes - but we have tools for this), but I think that's probably the best way forward from here.
I took a look at this, but it's not a lot of fun.
We currently use utidylib to clean HTML. This one only supports Python 2.
We could move to tidylib (notably without the u), which uses newer versions of everything and exists for python3. But the Python 3 version is not available until debian stretch.
We'd also have to carefully examine the difference from using tidylib vs utidylib, and should probably do that as a separate step. I guess we'll have to start there.