Our cursor interface has no nextObject method, so there's no point in
having a rewind method. Its major effect is ensuring that
fetch/forEach/map return no documents if you've already called one of
them once. It's not clear why this is actually useful to anybody.
rewind is kept around as a no-op; if we later implement nextObject, we
can make rewind do something, but we still presumably would auto-rewind
before fetch/forEach/map.
In minimongo, remove the db_objects cache inside each cursor. The only
actual use of this cache was that if you called count multiple times, it
would return the same number without re-running the query, and you could
share the query work between N calls to count and one call to
fetch/forEach/map (but only one call! future calls would return
nothing!) While there's a minor performance hit from getting rid of
this cache, it should also use a little less memory, and enable use
cases like
{{#with someCursor}}
{{#if count}}
{{#each this}}
...
{{/each}}
{{/if}}
{{/with}}
which didn't work before because even the deps invalidation didn't
rewind the cursor.
Also, as a minor optimization, skip an EJSON.clone if there's a
projection, because projection functions are guaranteed to clone.
Fixes#2114
This reverts commit f91d870778.
It turns out that while we did implement "storing EJSON in MongoDB", the
technique we chose is very flaky. Some MongoDB commands (insert,
updates that don't increase the doc size?) let you store fields with
non-starting $, and others (update that does increase the doc size?)
don't.
# Please enter a commit message to explain why this merge is necessary,
# especially if it merges an updated upstream into a topic branch.
#
# Lines starting with '#' will be ignored, and an empty message aborts
# the commit.
