01 Apr 2023: Searching my email attachments with Ponymail and Tika

Moving my email archives to Ponymail went well

One feature I forgot that zoe had was how it indexed some attachments.  If you have an email with a PDF attachment and that PDF attachment had plain text in it (it wasn't just a scan) then you could search on words in the PDF.  That's super handy.  Ponymail doesn't do that, in fact you don't get to search on any text in attachments, even if they are plain text (or things like patches).  Let's fix that!

Remember how I said whenever I need some code I first look if there is an Apache community that has a project that does something similar?  Well Apache Tika is an awesome project that will return the plain text of pretty much whatever you throw at it.  PDF? sure.  Patches? definitely.  Word docs? yup.  Images? yes.  Wait, images? so Tika will go and use Tesseract and do an OCR of an image.

Okay, so let's add a field to the mbox index, attachmenttext, populate it with Tika, and search on it.  For now if some text in an attachment matches your search query you'll see the result, but you won't know exactly where the text appears (perhaps later it could highlight which attachment it appears in).

I wrote a quick Python script that runs through all emails in Ponymail (or some search query subset), and if they have attachments runs all the attachments through Apache Tika, storing the plain texts in the attachmenttext field.  We ignore anything that's already got something in that field, so we can just run this periodically rather than on import.  Then a one-line patch to Ponymail also searches the attachmenttext field.  40,000 attachments and two hours later, it was all done and working.

It's not ready for a PR yet; probably for Ponymail upstream we'd want the option of doing this at import, although I chose not too so we can be deliberately careful as parsing untrusted attachments is risky

So there we have it; a way to search your emails including inside most attachments, outside the cloud, using Open Source projects and a few little patches.

Created: 01 Apr 2023
Tagged as: apache, fedora