fix: close the motor client if an invalid db name raises in __init__ (mongo-9)

self._client[database] can raise InvalidName after the client is already
built, leaking it via any exception traceback the caller holds. close it
before re-raising, matching the __aenter__ leak fix. also compresses essay
docstrings/comments across the module with no behavior change, keeping the
swallow-and-default contract note and the matched_count/nInserted footguns.

Signed-off-by: disqualifier <dev@disqualifier.me>

README.md

@@ -8,18 +8,18 @@ helpers for the common paths, with a raw escape hatch for everything else.
 `requirements.txt`:
 
 ```
-mongo @ git+ssh://git@git.rethinkstudios.io/rethink-public/mongo.git@v0.1.7
+mongo @ git+ssh://git@git.rethinkstudios.io/rethink-public/mongo.git@v0.1.8
 ```
 
 Direct:
 
 ```bash
-pip install "mongo @ git+ssh://git@git.rethinkstudios.io/rethink-public/mongo.git@v0.1.7"
+pip install "mongo @ git+ssh://git@git.rethinkstudios.io/rethink-public/mongo.git@v0.1.8"
 ```
 
 Requires `motor` and `pymongo` (pulled transitively).
 
-Drop the `@v0.1.7` suffix from the line above to install the latest unpinned.
+Drop the `@v0.1.8` suffix from the line above to install the latest unpinned.
 
 ## Usage
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "mongo"
-version = "0.1.7"
+version = "0.1.8"
 description = "async mongodb wrapper over motor with a raw escape hatch"
 requires-python = ">=3.10"
 dependencies = [

src/mongo/__init__.py

@@ -7,11 +7,10 @@ def __getattr__(name: str):
     """proxy bare package attribute access to the default instance (PEP 562)
 
     lets `import mongo; await mongo.get_documents(...)` work after init().
-    `from mongo import func` still won't see this (resolved before init).
+    `from mongo import func` still won't see this (resolved before init). the
-    note: the bare-proxy raw escape hatch is `mongo.collection(name)`, not
+    bare-proxy raw escape hatch is `mongo.collection(name)`, not `mongo[name]` —
-    `mongo[name]` — module-level subscripting isn't a thing in python, so the proxy
+    module-level subscripting isn't a thing in python, so the proxy only forwards
-    can only forward named attributes. on a Mongo instance both `db[name]` and
+    named attributes (an instance's `db[name]` still works fine).
-    `db.collection(name)` work.
     """
     if not name.startswith("_") and hasattr(MongoDB, name):
         return getattr(instance(), name)

src/mongo/mongo.py

@@ -12,8 +12,7 @@ object (preferred), one client per process:
     async with MongoDB(conn_string, database) as db:
         await db.get_documents("users", {})
 
-`Mongo` remains a back-compat alias of `MongoDB` — existing `Mongo(...)` call sites keep
+`Mongo` is a back-compat alias of `MongoDB`; old `Mongo(...)` call sites still work.
-working unchanged.
 
 module proxy (back-compat), arm once then call bare:
     import mongo                              # not `from mongo import ...`
@@ -21,12 +20,11 @@ module proxy (back-compat), arm once then call bare:
     await mongo.get_documents("users", {"active": True})
 
 errors:
-    wrapped methods log and swallow, returning a safe default
+    wrapped methods log and swallow, returning a safe default (False / [] / {} / 0 /
-    (False / [] / {} / 0 / None). .collection(name) / db[name] return
+    None). .collection(name) / db[name] return the raw motor collection: full driver
-    the raw motor collection: full driver surface, raises, nothing swallowed.
+    surface, raises, nothing swallowed. mongo SWALLOWS by design — the one deliberate
-    NOTE: mongo SWALLOWS by design — this is the one deliberate difference from the
+    difference from the fail-loud redis/psql/mysql trio, kept so existing consumers'
-    newer datastore trio (redis/psql/mysql), which is fail-loud. mongo's contract is
+    branch-on-result control flow doesn't break.
-    kept as-is so existing consumers' branch-on-result control flow doesn't break.
 
 naming consistency with the trio (all additive — old names still work):
     - class is `MongoDB` (was `Mongo`, kept as alias)
@@ -54,23 +52,26 @@ class MongoDB:
     """async mongodb wrapper; one client per process, attach to bot as bot.db"""
 
     def __init__(self, connection_string: str, database: str):
-        # motor builds the client eagerly here (no network I/O yet). a syntactically
+        # motor builds the client eagerly (no I/O yet); a bad URI raises InvalidURI
-        # malformed URI (e.g. "mongodb://") makes the driver raise InvalidURI at
+        # here with nothing built. an invalid db name raises below, after the client
-        # construction, before connect() — a deploy-time value, so it surfaces at startup
+        # exists — close it before propagating so it doesn't outlive the raise.
-        # either way and there is nothing to clean up.
         self._client = AsyncIOMotorClient(connection_string)
-        self._db = self._client[database]
+        try:
+            self._db = self._client[database]
+        except BaseException:
+            self._client.close()
+            raise
 
     async def connect(self) -> "MongoDB":
         """validate the connection with a ping and return self
 
-        motor connects lazily, so this is optional — call it to fail early on a bad
+        optional — motor connects lazily, so this just fails early on a bad
-        URI/credentials rather than on the first real op (parallel to the trio's
+        URI/credentials (parallel to the trio's connect()). raises, unlike the
-        connect()). raises on a bad connection, unlike the swallowing wrapped methods.
+        swallowing wrapped methods.
 
-        an instance is single-use: pymongo 4.x `close()` is irreversible, and connect()
+        an instance is single-use: pymongo 4.x `close()` is irreversible and
-        only pings (it does not rebuild the client). do not reuse an instance after
+        connect() doesn't rebuild the client. construct a fresh MongoDB instead of
-        close()/`async with` exit — construct a fresh MongoDB.
+        reusing one after close()/`async with` exit.
         """
         await self._client.admin.command("ping")
         return self
@@ -79,10 +80,9 @@ class MongoDB:
         try:
             return await self.connect()
         except BaseException:
-            # connect()/ping failing here would otherwise leak the motor client (built
+            # __aexit__ is not called when __aenter__ raises, so a failed ping here
-            # eagerly in __init__ with a background topology monitor + pool) — __aexit__
+            # would otherwise leak the motor client built in __init__. close it
-            # is not called when __aenter__ raises. close it before propagating so a
+            # before propagating.
-            # retry loop against a flapping server doesn't accumulate live clients.
             self.close()
             raise
 
@@ -122,9 +122,7 @@ class MongoDB:
         """create a collection, optionally seeding ONE index
 
         `index` is a single index key-spec (e.g. `[("a", 1), ("b", 1)]` builds one
-        compound index, NOT two). for multiple indexes call `create_index` per index.
+        compound index, NOT two); call `create_index` per index for multiples.
-        renamed from `indexes` in v0.1.6 to reflect that it seeds a single index; pass it
-        positionally, or update a keyword call site from `indexes=` to `index=`.
         """
         try:
             await self._db.create_collection(collection)
@@ -188,11 +186,10 @@ class MongoDB:
     ) -> int:
         """insert many documents, returning the inserted count
 
-        on a partial failure (e.g. ordered=False with one duplicate key among
+        on a partial failure (e.g. ordered=False with one duplicate key) the driver
-        several documents) the driver raises BulkWriteError AFTER inserting the
+        raises BulkWriteError AFTER inserting the rest; count comes from
-        non-failing documents; that count is read from `e.details['nInserted']`
+        `e.details['nInserted']` so a caller retrying on a wrongly-reported 0 doesn't
-        so the return value stays honest — a caller retrying the whole batch on a
+        re-insert documents that already landed.
-        wrongly-reported 0 would re-insert documents that already landed.
         """
         try:
             result = await self._db[collection].insert_many(documents, ordered=ordered)
@@ -379,9 +376,9 @@ class MongoDB:
     async def update_document_field(self, collection: str, target: dict, updates: dict) -> bool:
         """$set one or more fields on a single document
 
-        returns True when a document matched, even if the write changed nothing (a
+        returns True when a document matched, even if `$set` to an identical value
-        `$set` to the identical value leaves `modified_count=0`); use `matched_count`
+        leaves `modified_count=0` — uses `matched_count` so an idempotent no-op isn't
-        so an idempotent no-op on an existing doc isn't misread as a failure.
+        misread as a failure.
         """
         try:
             response = await self._db[collection].update_one(target, {"$set": updates})
@@ -395,9 +392,8 @@ class MongoDB:
     ) -> bool:
         """apply raw update operators ($set/$inc/$unset/...) to a single document
 
-        returns True when a document matched, even if the operators changed nothing
+        returns True when a document matched, even if the operators changed nothing —
-        (e.g. `$set` to an identical value) — use `matched_count` so an idempotent
+        uses `matched_count` so an idempotent write isn't misread as a failure.
-        write on an existing doc isn't misread as a failure.
         """
         try:
             response = await self._db[collection].update_one(target, update_query)
@@ -409,10 +405,9 @@ class MongoDB:
     async def upsert_document_field(self, collection: str, target: dict, updates: dict) -> bool:
         """$set fields on a single document, creating it if absent
 
-        returns True when a document matched or was upserted, even if the write
+        returns True when a document matched or was upserted, even if `$set` to an
-        changed nothing (a `$set` to the identical value leaves `modified_count=0`);
+        identical value leaves `modified_count=0` — mirrors update_document_field's
-        use `matched_count` so an idempotent no-op upsert on an existing doc isn't
+        matched_count-over-modified_count contract.
-        misread as a failure — mirrors update_document_field/update_document_operator.
         """
         try:
             response = await self._db[collection].update_one(
@@ -439,9 +434,8 @@ class MongoDB:
     ) -> bool:
         """$push a value onto an array field
 
-        returns True when a document matched (consistent with the other single-doc
+        returns True when a document matched; $push always mutates, so matched_count
-        update helpers); $push always mutates, so matched_count and modified_count
+        and modified_count agree here in practice.
-        agree here in practice.
         """
         try:
             response = await self._db[collection].update_one(target, {"$push": {array: value}})
@@ -456,8 +450,7 @@ class MongoDB:
     ) -> bool:
         """$push to an array and $set a field in one update
 
-        returns True when a document matched (consistent with the other single-doc
+        returns True when a document matched (matched_count, not modified_count).
-        update helpers).
         """
         try:
             response = await self._db[collection].update_one(
@@ -474,9 +467,8 @@ class MongoDB:
     ) -> bool:
         """$pull a value from an array field
 
-        returns True when a document matched, even if nothing was pulled (the value
+        returns True when a document matched, even if the value was absent and
-        was absent) — use `matched_count` so a no-op pull on an existing doc isn't
+        nothing was pulled — uses `matched_count`, not `modified_count`.
-        misread as a failure.
         """
         try:
             response = await self._db[collection].update_one(target, {"$pull": {array: value}})
@@ -556,10 +548,9 @@ class MongoDB:
     #   bulk / checks
 
     async def bulk_write(self, collection: str, operations: list, ordered: bool = True) -> dict:
-        """
+        """run a batch of pymongo write ops (InsertOne/UpdateOne/DeleteOne/...)
-        run a batch of pymongo write ops (InsertOne/UpdateOne/DeleteOne/...)
 
-        caller builds the ops from pymongo, e.g.:
+        caller builds the ops, e.g.:
             from pymongo import UpdateOne
             ops = [UpdateOne({'_id': i}, {'$set': {...}}, upsert=True) for i in ids]
         returns a summary dict of counts, or an empty dict on failure
@@ -580,8 +571,7 @@ class MongoDB:
     async def document_check_multi(
         self, collection: str, target: dict, checks: List[str], value: str
     ) -> bool:
-        """
+        """return whether a doc matching target has value in any of the given arrays
-        return whether a doc matching target has value in any of the given arrays
 
         target is a normal filter dict (merged into the query), not a bare _id
         """
@@ -602,12 +592,9 @@ Mongo = MongoDB
 
 
 # -----------------------------------------------------------------------------
-#   backwards-compat module proxy
+#   backwards-compat module proxy: `import mongo; mongo.init(conn, db)` then
-#       - lets legacy call sites keep using `await mongo.get_documents(...)`
+#   bare `await mongo.func(...)`. needs `import mongo` — `from mongo import func`
-#         without an object, after a one-time `mongo.init(conn, db)`
+#   resolves at import time, before init() runs.
-#       - works with `import mongo; mongo.func(...)` (resolved at call time)
-#       - does NOT work with `from mongo import func` (resolved at import,
-#         before init runs) — switch those sites to `import mongo`
 
 _default: Optional[MongoDB] = None