I understand that I have already been given an answer here, but I am still curious as to why this is the case (perhaps I should ask this on the hackers list though, if so let me know).
More importantly I'd like to understand why I would need to use the start/stop backup commands to ensure a valid backup when using filesystem snapshots (assuming I get the order correct)- worst case scenario wouldn't it be the same as a crash and cause an automatic roll-forward?
pg_start_backup('backup_label') and pg_stop_backup(), if I understand it correctly, write to the 'backup_label' file the information necessary to recover "consistently" from that backup. For example, backup_label file contains the checkpoint location and its REDO location (identified as "START WAL LOCATION:" field in the backup_label file.) While you are reading the code, you can read the comment above the function read_backup_label() in src/backend/access/transam/xlog.c