Add technote about avoiding feed parsing.

Technotes/AvoidFeedParsing.markdown

@@ -0,0 +1,39 @@
+# How NetNewsWire Avoids Parsing Feeds
+
+NetNewsWire’s code for reading feeds directly — not via a syncing system like Feedbin or Feedly) — does its best to avoid parsing feeds.
+
+Here’s the thing: parsing a feed means not just parsing the feed, it also means comparing the parsed version to what’s in the database. This is all *work*, and part of performing well is to avoid work.
+
+Here’s what it does:
+
+## Conditional GET
+
+I can’t stress this strongly enough. When downloading a feed, NetNewsWire sends the appropriate headers to give the server a chance to respond with 304 Not Modified.
+
+It’s quite simple — read The Fishbowl’s [HTTP Conditional Get for RSS Hackers](https://fishbowl.pastiche.org/2002/10/21/http_conditional_get_for_rss_hackers) (from way back in 2002!) for how this works.
+
+This is such a great thing! It means less bandwidth uses, less energy consumed, etc.
+
+Unfortunately, not every server implements the server side of Conditional GET. (Boo.) So we have a second method of avoiding work.
+
+## Feed Content Hashing
+
+When NetNewsWire parses a feed, it creates a hash — MD5 is fine for this sort of thing — of the content of the feed.
+
+It stores that hash along with other feed metadata.
+
+The next time it downloads the feed, it generates a hash of the just-downloaded copy. If the new hash matches the old hash, then the feed hasn’t changed, and we skip parsing it. Yay!
+
+## Additional Fallbacks
+
+NetNewsWire also looks at the content of the feed. If it’s definitely an image and not an RSS feed, for instance, it doesn’t attempt to parse it.
+
+Yes, this kind of thing happens in the real world: I’ve seen it. (Once I even saw a feed URL return a movie file.)
+
+We could more here, but it’s not often an issue, so it’s not a high priority. Just a good-to-have.
+
+## Thing It Never Does
+
+Feeds sometimes contain dates for modification times. NetNewsWire doesn’t trust these at all. In-feed dates are *never* used for making any decisions about parsing or not.
+
+When an article has a modification date, that date is stored in the database. But it’s there only in case it should be shown to the user. (Sometimes articles in a feed have a modification date but not a publication date — why oh why? — and in that case we display the modification date.)