Revert "Always scan for media changes"

This reverts commit 09cb8b3cf6.

Overhead on larger folders/slower devices is more than I originally
anticipated, and can run into multiple seconds. This seems to be
particularly egregious on mobile, which I presume is due to sandboxing
overhead.

rslib/src/media/changetracker.rs

@@ -9,7 +9,8 @@ use crate::{
     media::{
         database::{MediaDatabaseContext, MediaEntry},
         files::{
-            filename_if_normalized, sha1_of_file, MEDIA_SYNC_FILESIZE_LIMIT, NONSYNCABLE_FILENAME,
+            filename_if_normalized, mtime_as_i64, sha1_of_file, MEDIA_SYNC_FILESIZE_LIMIT,
+            NONSYNCABLE_FILENAME,
         },
     },
     prelude::*,
@@ -59,7 +60,17 @@ where
 
     pub(super) fn register_changes(&mut self, ctx: &mut MediaDatabaseContext) -> Result<()> {
         ctx.transact(|ctx| {
-            debug!(self.log, "begin change check");
+            // folder mtime unchanged?
+            let dirmod = mtime_as_i64(self.media_folder)?;
+
+            let mut meta = ctx.get_meta()?;
+            debug!(self.log, "begin change check"; "folder_mod" => dirmod, "db_mod" => meta.folder_mtime);
+            if dirmod == meta.folder_mtime {
+                debug!(self.log, "skip check");
+                return Ok(());
+            } else {
+                meta.folder_mtime = dirmod;
+            }
 
             let mtimes = ctx.all_mtimes()?;
             self.checked += mtimes.len();
@@ -70,6 +81,8 @@ where
             self.add_updated_entries(ctx, changed)?;
             self.remove_deleted_files(ctx, removed)?;
 
+            ctx.set_meta(&meta)?;
+
             // unconditional fire at end of op for accurate counts
             self.fire_progress_cb()?;
 

rslib/src/media/database.rs

@@ -56,10 +56,6 @@ pub struct MediaEntry {
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct MediaDatabaseMetadata {
-    /// The syncing code no longer uses this; files are scanned for
-    /// indiscriminately. After this has been in production for a while
-    /// without reports of speed regressions, we should remove the rest
-    /// of the code that refers to this.
     pub folder_mtime: i64,
     pub last_sync_usn: i32,
 }