Collection names are allowed to have a '.' in their name, db
names aren't (see
https://www.mongodb.com/docs/manual/reference/limits/#naming-restrictions)
This codes changes the logic to extract the collection
name form a FullConnecton string by stripping until the first
dot, and provides a test case.
780df80cfc introduces support for the
OP_MSG protocol. Unfortunately, the upsert and multi options of the
update command still use flagBits to communicate the options, whereas
they must be provided directly into the command document,
alongside the "q" and "v" fields.
This commit:
- introduces a test for a single-document upsert that, if isolated,
succeeds against the reference MongoDB 3.6 container, but fails
against an official 6.0 image.
- provides a patch that sets the appropriate options.
The test is not perfect as the upsert operation is inherently racy and
this difficult to test. A comfortable threadDelay has been inserted as
a workaround to accomodate for medium workloads.
Updating to avoid deprecated Producer/Consumer from `conduit`.
Removed unused imports.
Removing superfluous brackets.
Simplifying a few function bodies with catMaybe, fromMaybe, mapMaybe.
Some new users might be put off by the lack of a array of choice between connection methods
that this package offers. This remark and the associated examples are a first attempt to make the package easier to start with.
--
The `newPipeline` function, used as part of `connect`, forks a
listener thread. Before this commit, the thread is forked with
`forkFinally`, where the thread action is run in the same mask as the
parent thread. The thread is then killed by a `killThread` when
closing a connection.
This is typically not a problem if the mask is “masked” (or,
obviously, “unmasked”), because the listener is generally blocked on a
channel at some time or other, and therefore will accept the
asynchronous exception thrown by `killThread`, and terminate.
However, if the mask is “masked uninterruptible”, then the listener
definitely doesn't receive asynchronous exceptions, and the
`killThread` calls hangs, and never returns.
One should probably never call `connect` in a “masked uninterruptible”
action. However, it sounds better to protect the mongoDB library
against the user accidentally doing so than to add a big warning
saying that calling `connect` in “masked uninterruptible” will cause
the program to hang down the line.
Therefore, this commit uses `forkIOWithUnmask`, in order to run the
thread action always in an “unmasked” state. In which case we can be
sure that we can always kill the listener thread regardless of the
client code.
In particular this allows you to use `fromIntegral` without having to add the newtype wrapper. This can help existing code move away from importing and referencing the PortID type altogether.
Add flag imitating bson package PR for network changes.
Add stack files for compilation checking.
Both ghc86 builds work. Still need to fix ghc84 and under builds with older network code.
This reverts commit 5f04dc6a23.
Leave the use of Producer and Consumer for now until we drop support for conduit-1.2.*.
conduit-1.3 introduces ConduitT and deprecates the use of type synonyms.
However, ConduitT is not present in conduit-1.2.