The notes/amlog database is used to annotate individual commits that
result from "git am" application. A note is attached to the commit
object and record s the Message-ID of the incoming e-mailed patch
that resulted in the commit, so "git show --notes=amlog" would
easily show where the commit came from. The rewrite-hook mechanism
can be used to maintain the notes across rebasing and amending (but
cherry-pick does not preserve the note by design---the maintainer
has to be careful to avoid using cherry-pick). One message can and
does result in multiple commits, and the mapping worked naturally in
this direction.
Originally it felt like a good idea to create a blob object that has
a Message-ID in it, and annotate the blob object with a note message
that has the name of the commit object that results by running "git
am" on the message, and mix such records in the notes database.
When you have a message, from the Message-ID, you can manufacture a
blob that has the Message-ID in it and ask the notes database about
the note attached to it, effectively giving you a reverse mapping.
This was ugly, unnecessary and unworkable at the same time.
- These blobs with message-ID in them are not anchored by any ref;
the reverse mapping entries in the notes tree were subject to be
gc'ed any time.
- "git grep -e '<message-id>' notes/amlog" essentially gives a
mechanism that is quick enough to find what commits resulted from
a message.
- There is no machinery that helps these reverse mapping notes to
be maintained across rebases and amends. Because a single "git
am" session is often far from enough for an e-mailed patch, these
"reverse" entries that were created upon the first application
quickly became stale pointing at commits that have been amended
away. There often are more than one commit that result from the
same message on topic branches while the right base is being
selected, so "the last one wins" rule, if it were even possible
to implement, wouldn't have been sufficient.
Since grepping for the Message-ID in the notes database, i.e. using
the forward mapping, gives an access to the necessary piece of
information reliably and quicly enough, let's retire the failed
attempt to throw reverse mapping entries in to the "amlog" notes.