quantpad-data 0.7.1__tar.gz → 0.8.1__tar.gz

{quantpad_data-0.7.1 → quantpad_data-0.8.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: quantpad-data
-Version: 0.7.1
+Version: 0.8.1
 Summary: Official Python client for the QuantPad market-data API
 Project-URL: Documentation, https://api.quantpad.ai/external/docs
 Author: QuantPad
@@ -106,9 +106,8 @@ symbol it is. That is convenient until it is wrong, and when it is wrong it
 fails quietly — asking for a specific futures contract returns an empty frame
 that looks exactly like a coverage gap.
 
-`qpd.v2.Historical` removes the inference. It takes the same arguments as
-`databento.Historical`, so code written against Databento ports over by
-swapping the client:
+`qpd.v2.Historical` removes the inference. Every call names its dataset and
+symbology type explicitly, and every response reports the address it used:
 
 ```python
 import quantpad_data as qpd
@@ -243,7 +242,11 @@ cache.clear()         # throw it all away
 (`list_datasets`, `list_schemas`, `list_fields`, `list_publishers`,
 `get_dataset_range`, `get_dataset_condition`, `get_record_count`), and
 `client.options.chain(...)` for enumerating strikes and expiries without
-downloading a day's entire `definition` schema.
+downloading a day's entire `definition` schema. A chain is capped at `limit`
+(5,000 by default, 50,000 at most); the frame's `attrs` carry the server's
+`count`, `total` and `truncated`, and a cut chain raises
+`TruncatedChainWarning` — `warnings.simplefilter("error",
+qpd.TruncatedChainWarning)` makes a partial chain a hard failure.
 
 There is no cost or unit-price surface. QuantPad data is included in the plan
 rather than metered per request.

{quantpad_data-0.7.1 → quantpad_data-0.8.1}/README.md

@@ -84,9 +84,8 @@ symbol it is. That is convenient until it is wrong, and when it is wrong it
 fails quietly — asking for a specific futures contract returns an empty frame
 that looks exactly like a coverage gap.
 
-`qpd.v2.Historical` removes the inference. It takes the same arguments as
-`databento.Historical`, so code written against Databento ports over by
-swapping the client:
+`qpd.v2.Historical` removes the inference. Every call names its dataset and
+symbology type explicitly, and every response reports the address it used:
 
 ```python
 import quantpad_data as qpd
@@ -221,7 +220,11 @@ cache.clear()         # throw it all away
 (`list_datasets`, `list_schemas`, `list_fields`, `list_publishers`,
 `get_dataset_range`, `get_dataset_condition`, `get_record_count`), and
 `client.options.chain(...)` for enumerating strikes and expiries without
-downloading a day's entire `definition` schema.
+downloading a day's entire `definition` schema. A chain is capped at `limit`
+(5,000 by default, 50,000 at most); the frame's `attrs` carry the server's
+`count`, `total` and `truncated`, and a cut chain raises
+`TruncatedChainWarning` — `warnings.simplefilter("error",
+qpd.TruncatedChainWarning)` makes a partial chain a hard failure.
 
 There is no cost or unit-price surface. QuantPad data is included in the plan
 rather than metered per request.

{quantpad_data-0.7.1 → quantpad_data-0.8.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "quantpad-data"
-version = "0.7.1"
+version = "0.8.1"
 description = "Official Python client for the QuantPad market-data API"
 readme = "README.md"
 requires-python = ">=3.10"

{quantpad_data-0.7.1 → quantpad_data-0.8.1}/quantpad_data/__init__.py

@@ -198,18 +198,22 @@ def iter_l1_mbp1_df_chunks(symbol, start_ms, end_ms, chunksize=250_000,
 
 from . import cache  # noqa: E402  (submodule: on-disk cache for settled /v2)
 from . import lean  # noqa: E402  (submodule: LEAN backtesting data staging)
-from . import v2  # noqa: E402  (submodule: Databento-shaped explicit API)
+from . import v2  # noqa: E402  (submodule: explicit-address API)
 
 # Exported at the root as well as on ``v2``: the sandbox module reaches it
 # as ``qpd.Historical()``, and one spelling across both surfaces means
 # notebook code moves between them unchanged.
 Historical = v2.Historical
+# At the root for the same reason: ``warnings.simplefilter("error",
+# qpd.TruncatedChainWarning)`` is how a script turns a cut chain into a
+# failure, and it should not need to know which submodule warns.
+TruncatedChainWarning = v2.TruncatedChainWarning
 
 __all__ = [
     "Client", "QuantPadClient", "QuantPadDataError", "QuantPadDataNotReady", "AuthenticationError",
     "PermissionDeniedError", "RateLimitError", "QuotaExceededError",
     "NotFoundError", "V2NotAvailableError", "ValidationError", "UpstreamError",
-    "ResponseTooLargeError", "TICK_SCHEMAS",
+    "ResponseTooLargeError", "TruncatedChainWarning", "TICK_SCHEMAS",
     "ROLL_ADJUST_MODES", "get_bars", "get_bars_with_retry", "get_ticks",
     "get_trades", "get_mbp1", "get_mbp10", "get_trades_rth", "get_mbp1_rth",
     "get_mbp10_rth", "get_spread_1m_rth", "symbology_resolve", "get_universe",

{quantpad_data-0.7.1 → quantpad_data-0.8.1}/quantpad_data/client.py

@@ -84,7 +84,7 @@ V2_NOT_AVAILABLE = (
     "QuantPad support about /v2 access."
 )
 
-VERSION = "0.7.1"
+VERSION = "0.8.1"
 USER_AGENT = f"quantpad-data-python/{VERSION}"
 TICK_SCHEMAS = (
     "trades", "mbp-1", "mbp-10", "cmbp-1", "tcbbo", "cbbo-1s",

{quantpad_data-0.7.1 → quantpad_data-0.8.1}/quantpad_data/memory.py

@@ -15,7 +15,7 @@ before fetching it, which is why it takes both.
 
 THE UNITS ARE NOT THE SAME, AND THAT IS THE WHOLE DIFFICULTY
 The service prices in DBN bytes: record count times record width, which is
-Databento's own billing arithmetic.  This process pays for the compressed wire
+upstream's own billing arithmetic.  This process pays for the compressed wire
 bytes, an Arrow table decoded from them, and a pandas frame copied out of that
 table -- partly all three at once.  Measured with
 scripts/measure_client_memory.py against production, peak RSS per priced byte

{quantpad_data-0.7.1 → quantpad_data-0.8.1}/quantpad_data/v2.py

@@ -1,4 +1,4 @@
-"""Databento-shaped access to QuantPad's cache.
+"""Explicit-address access to QuantPad's cache.
 
 The v1 helpers take QuantPad's own arguments and infer the rest: which
 dataset a symbol lives in, what kind of symbol it is, how far back to
@@ -7,18 +7,7 @@ it is wrong it fails quietly — a specific futures contract returns an
 empty frame that looks exactly like a coverage gap.
 
 This module removes the inference.  Every call names its dataset and its
-symbology type, exactly as ``databento.Historical`` requires, and every
-response reports the address it actually used.  Code written against the
-Databento client transfers over with the client swapped:
-
-    import databento as db
-    client = db.Historical(key=...)
-    df = client.timeseries.get_range(
-        dataset="GLBX.MDP3", symbols="ESZ6", schema="ohlcv-1m",
-        start="2026-01-05", end="2026-01-06",
-    ).to_df()
-
-becomes
+symbology type, and every response reports the address it actually used:
 
     import quantpad_data as qpd
     client = qpd.v2.Historical()
@@ -27,9 +16,9 @@ becomes
         start="2026-01-05", end="2026-01-06",
     )
 
-The differences are deliberate and few: a DataFrame comes back directly
-rather than a DBNStore, and there is no cost or billing surface, because
-data is included in the QuantPad plan rather than metered per request.
+A DataFrame comes back directly, and there is no cost or billing surface,
+because data is included in the QuantPad plan rather than metered per
+request.
 """
 
 from __future__ import annotations
@@ -38,6 +27,7 @@ import io
 import json
 import os
 import tempfile
+import warnings
 from datetime import date, timedelta
 from typing import Any, Iterator, Sequence
 
@@ -254,7 +244,7 @@ class _Namespace:
 
 
 class Timeseries(_Namespace):
-    """Mirrors ``databento.Historical.timeseries``."""
+    """Ranges of bars, ticks and book depth for an explicit address."""
 
     def get_range(
         self,
@@ -762,12 +752,12 @@ class Timeseries(_Namespace):
 
 
 class Symbology(_Namespace):
-    """Mirrors ``databento.Historical.symbology``, plus a way in.
+    """Symbol resolution, plus a way in.
 
     Every other call here demands an explicit ``dataset``, which is the
     point and also a dead end for someone holding the string "AAPL".
-    ``search`` is the one endpoint that answers that, and it has no
-    upstream counterpart because Databento assumes you already know.
+    ``search`` is the one endpoint that answers that; the rest of this
+    surface assumes you already know.
     """
 
     def search(
@@ -829,7 +819,7 @@ class Symbology(_Namespace):
 
 
 class Metadata(_Namespace):
-    """Mirrors ``databento.Historical.metadata``.
+    """Datasets, schemas, fields, coverage and request sizing.
 
     The cost and unit-price endpoints are deliberately absent: QuantPad
     data is included in the plan rather than metered, so a per-request
@@ -920,13 +910,61 @@ class Metadata(_Namespace):
         return self._client._request("GET", path, params=clean).json()
 
 
+class TruncatedChainWarning(UserWarning):
+    """``options.chain`` returned fewer contracts than the filters matched.
+
+    The service caps a chain at ``limit`` (5,000 by default, 50,000 at
+    most) and says so in the body; this is that flag made audible.  Raise
+    ``limit`` or narrow the filters (``expiration``, ``strike_min``/
+    ``strike_max``, ``instrument_class``) for the whole chain.
+    """
+
+
+def _chain_frame(payload: dict) -> pd.DataFrame:
+    """The chain as a frame, carrying the server's count beside the rows.
+
+    The route answers ``count`` (rows in the body), ``total`` (rows the
+    filters matched) and ``truncated`` beside ``contracts``.  Until this
+    the frame was built from ``contracts`` alone and the other three were
+    dropped, so a notebook handed exactly 5,000 rows could not tell a cut
+    chain from a whole one -- verified on prod, where an unfiltered
+    SPY.OPT came back as exactly the default limit against ~12,470
+    instruments.  The three ride on ``frame.attrs`` (the same place
+    ``get_range`` keeps its echo) and a truncation is also warned, since
+    ``attrs`` is invisible to anyone not already looking for it.
+
+    Older servers send no ``total``; it is then unknown rather than
+    guessed, and ``truncated`` is read as the server sent it.
+    """
+    contracts = payload.get("contracts", [])
+    frame = pd.DataFrame(contracts)
+    truncated = bool(payload.get("truncated", False))
+    frame.attrs["count"] = payload.get("count", len(contracts))
+    frame.attrs["total"] = payload.get("total")
+    frame.attrs["truncated"] = truncated
+    if truncated:
+        total = frame.attrs["total"]
+        matched = f"{total:,} matched" if isinstance(total, int) else "more matched"
+        warnings.warn(
+            f"options.chain returned {len(contracts):,} contracts but "
+            f"{matched}: the chain was cut at the service's limit. Raise "
+            "limit= (up to 50,000) or narrow the filters to see the rest.",
+            TruncatedChainWarning,
+            stacklevel=3,
+        )
+    return frame
+
+
 class Options(_Namespace):
     """Options-chain enumeration.
 
-    Not part of Databento's client surface — upstream expects you to pull
-    the whole ``definition`` schema and filter it yourself.  This does
+    Not part of the usual historical-client surface, which expects you to
+    pull the whole ``definition`` schema and filter it yourself.  This does
     that server-side, because "download every instrument definition for
     the day" is a poor answer to "what strikes exist".
+
+    The frame's ``attrs`` carry ``count``, ``total`` and ``truncated``;
+    a truncated chain also raises :class:`TruncatedChainWarning`.
     """
 
     def chain(
@@ -955,11 +993,11 @@ class Options(_Namespace):
         payload = self._client._request(
             "GET", "/v2/options/chain", params=clean
         ).json()
-        return pd.DataFrame(payload.get("contracts", []))
+        return _chain_frame(payload)
 
 
 class Batch(_Namespace):
-    """Mirrors ``databento.Historical.batch``, scoped to your own jobs.
+    """Batch extracts, scoped to your own jobs.
 
     Useful when a range is too large to stream in one sitting: submit,
     walk away, then hydrate the result into the cache so ordinary
@@ -1010,7 +1048,7 @@ class Batch(_Namespace):
 
 
 class Historical:
-    """A ``databento.Historical``-shaped view of QuantPad's cache.
+    """An explicit-address view of QuantPad's cache.
 
     Takes the same credentials as the v1 ``Client`` — an API key, or the
     injected token when running inside a QuantPad sandbox.

{quantpad_data-0.7.1 → quantpad_data-0.8.1}/tests/test_packaging.py

@@ -21,6 +21,7 @@ from __future__ import annotations
 
 import ast
 import os
+import re
 import subprocess
 import sys
 import tarfile
@@ -53,6 +54,7 @@ def built(tmp_path_factory):
         # ./dist would make it skip in CI, where nothing has built there yet --
         # a test that skips exactly where it is needed.
         "wheel_path": wheel,
+        "sdist_path": sdist,
     }
 
 
@@ -182,3 +184,39 @@ class TestAFreshInstallExposesTheApi:
             f"a fresh install cannot answer for its own API:\n"
             f"{done.stdout}\n{done.stderr[-1500:]}"
         )
+
+
+class TestNothingShippedNamesTheDataVendor:
+    """QuantPad presents its market data as its own, and a published package is
+    as public as a response: PyPI serves every file in both artifacts, the
+    README is the project page, and ``help()`` in a user's sandbox prints the
+    docstrings. So nothing shipped may name the vendor -- read out of the
+    artifacts, for the same reason as everything else in this file."""
+
+    VENDOR = re.compile(r"data[\s_-]?bento", re.IGNORECASE)
+
+    @staticmethod
+    def _texts(built, kind):
+        if kind == "wheel":
+            with zipfile.ZipFile(built["wheel_path"]) as archive:
+                for name in archive.namelist():
+                    yield name, archive.read(name)
+        else:
+            with tarfile.open(built["sdist_path"]) as archive:
+                for member in archive.getmembers():
+                    if member.isfile():
+                        yield member.name, archive.extractfile(member).read()
+
+    @pytest.mark.parametrize("kind", ["wheel", "sdist"])
+    def test_no_shipped_file_names_the_vendor(self, built, kind):
+        offenders = [
+            name
+            for name, data in self._texts(built, kind)
+            if self.VENDOR.search(data.decode("utf-8", errors="replace"))
+        ]
+        assert not offenders, f"the {kind} names the data vendor in {offenders}"
+
+    def test_the_readme_is_in_what_was_checked(self, built):
+        """Guards the test above: the wheel's METADATA carries the README."""
+        names = [name for name, _ in self._texts(built, "wheel")]
+        assert any(name.endswith("METADATA") for name in names)

{quantpad_data-0.7.1 → quantpad_data-0.8.1}/tests/test_v2.py

@@ -1,4 +1,4 @@
-"""The Databento-shaped surface.
+"""The explicit-address surface.
 
 Two things are being checked. First that the wire calls carry exactly
 what the caller said and nothing the SDK invented — the whole reason
@@ -95,7 +95,7 @@ class TestNothingIsInferred:
         assert kwargs["json"]["symbols"] == ["ESZ6"]
 
     def test_a_single_symbol_string_is_sent_as_a_list(self):
-        """Databento accepts either; normalising here keeps the server
+        """Upstream accepts either; normalising here keeps the server
         contract one shape instead of two."""
         api, session = client_with([response(200, arrow_bytes(bars_table()))])
         api.timeseries.get_range(
@@ -104,7 +104,7 @@ class TestNothingIsInferred:
         )
         assert session.calls[0][2]["json"]["symbols"] == ["ESZ6"]
 
-    def test_the_default_stype_matches_databento_not_our_own_convention(self):
+    def test_the_default_stype_matches_upstream_not_our_own_convention(self):
         api, session = client_with([response(200, arrow_bytes(bars_table()))])
         api.timeseries.get_range(
             dataset="GLBX.MDP3", symbols="ESZ6", schema="ohlcv-1m",
@@ -2117,6 +2117,79 @@ class TestTheOtherNamespaces:
         assert "strike_min" not in params
         assert "instrument_class" not in params
 
+    # chaintruncate: the route says when it cut the chain, and the frame
+    # used to drop that.  Verified on prod: unfiltered SPY.OPT returned
+    # exactly 5,000 rows (the default limit) against ~12,470 instruments,
+    # with no way to tell from the frame.
+
+    def test_a_cut_chain_warns_and_carries_the_servers_count(self):
+        import warnings
+
+        api, _ = client_with([json_response({
+            "count": 2, "total": 12470, "truncated": True,
+            "contracts": [
+                {"raw_symbol": "SPY   260116C00500000"},
+                {"raw_symbol": "SPY   260116P00500000"},
+            ],
+        })])
+        with warnings.catch_warnings(record=True) as caught:
+            warnings.simplefilter("always")
+            chain = api.options.chain(
+                dataset="OPRA.PILLAR", underlier="SPY.OPT", date="2026-01-05",
+            )
+        assert len(chain) == 2
+        assert chain.attrs["truncated"] is True
+        assert chain.attrs["count"] == 2
+        assert chain.attrs["total"] == 12470
+        (warning,) = [
+            w for w in caught if issubclass(w.category, quantpad_data.TruncatedChainWarning)
+        ]
+        text = str(warning.message)
+        assert "2 contracts" in text and "12,470 matched" in text
+        assert "limit=" in text  # the way out is named, not just the fact
+
+    def test_a_whole_chain_is_silent_and_says_so(self):
+        import warnings
+
+        api, _ = client_with([json_response({
+            "count": 1, "total": 1, "truncated": False,
+            "contracts": [{"raw_symbol": "ESM6 C4800"}],
+        })])
+        with warnings.catch_warnings():
+            warnings.simplefilter("error")  # any warning here is a failure
+            chain = api.options.chain(
+                dataset="GLBX.MDP3", underlier="ES.OPT", date="2026-01-05",
+            )
+        assert chain.attrs["truncated"] is False
+        assert chain.attrs["total"] == 1
+
+    def test_an_older_server_without_total_still_reports_the_cut(self):
+        """The deployed route sent ``truncated`` before it sent ``total``;
+        a frame from it must still say it was cut, with the total unknown
+        rather than invented."""
+        import warnings
+
+        api, _ = client_with([json_response({
+            "count": 1, "truncated": True, "contracts": [{"raw_symbol": "A"}],
+        })])
+        with warnings.catch_warnings(record=True) as caught:
+            warnings.simplefilter("always")
+            chain = api.options.chain(
+                dataset="GLBX.MDP3", underlier="ES.OPT", date="2026-01-05",
+            )
+        assert chain.attrs["truncated"] is True
+        assert chain.attrs["total"] is None
+        assert any(
+            issubclass(w.category, quantpad_data.TruncatedChainWarning) for w in caught
+        )
+
+    def test_the_truncation_warning_is_exported_at_the_root(self):
+        """``warnings.simplefilter("error", quantpad_data.TruncatedChainWarning)`` is
+        how a script refuses to run on a partial chain; it needs the name
+        without knowing which submodule warns."""
+        assert quantpad_data.TruncatedChainWarning is quantpad_data.v2.TruncatedChainWarning
+        assert issubclass(quantpad_data.TruncatedChainWarning, UserWarning)
+
     def test_batch_submit_and_hydrate_are_reachable(self):
         api, session = client_with([
             json_response({"job_id": "j1"}),