Align native contract and SQL correctness#184
Conversation
nicosuave
force-pushed
the
review/python-native-contract
branch
from
6ccaf41 to
cd7a08e
Compare
nicosuave
changed the title
nicosuave
force-pushed
the
review/cli-ci-release-hardening
branch
from
326838a to
c6b38ca
Compare
nicosuave
force-pushed
the
review/python-native-contract
branch
from
cd7a08e to
4b9c638
Compare
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4b9c638ef7
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if jp.custom_condition: | ||
| join_cond = self._custom_join_condition(jp) |
There was a problem hiding this comment.
When a relationship sql uses columns that are not the declared foreign/primary keys (for example {from}.tenant_id = {to}.tenant_id AND {to}.valid_to IS NULL), this branch emits the predicate against the per-model CTE aliases, but _build_model_cte only projects dimensions, metrics, PKs, and declared join keys. The generated SQL then references missing CTE columns like customers_cte.valid_to; either extract/project columns from the custom predicate or reject predicates that use non-projected columns.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 38929833cf
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| METRIC_FIELDS = { | ||
| "name", | ||
| "extends", | ||
| "type", | ||
| "agg", | ||
| "sql", | ||
| "expr", | ||
| "measure", |
There was a problem hiding this comment.
With the new strict field validation, native YAML exported by previous versions for a derived metric can no longer be loaded: the old exporter wrote a metric-local metrics: [...] dependency list for derived metrics, but this allow-list omits that field, so _parse_metric() now raises unknown native field(s) before it can ignore or migrate it. This affects users re-loading previously exported Sidemantic YAML whenever a derived metric had detected dependencies; allow the legacy metrics key or handle it during parsing to keep those files compatible.
Useful? React with 👍 / 👎.
nicosuave
force-pushed
the
review/python-native-contract
branch
from
3892983 to
262e1a8
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 262e1a8f9a
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if resolved_metric and resolved_model_name is None: | ||
| metric_expr = self._build_metric_sql(resolved_metric) | ||
| metric_expr = self._wrap_with_fill_nulls(metric_expr, resolved_metric) | ||
| alias = aliases.get(metric_ref, resolved_metric.name) |
There was a problem hiding this comment.
When a graph-level metric itself contains a dot (now resolved before treating model.metric as a model measure), this branch emits the SELECT alias as the full metric name, e.g. AS "orders.revenue", but the existing ORDER BY path later strips everything before the dot and orders by revenue. A query like metrics=["orders.revenue"] with a graph-level metric named orders.revenue and order_by=["orders.revenue DESC"] therefore generates ORDER BY revenue DESC, which does not match the projected alias and fails unless an unrelated revenue column exists; use the same alias normalization for dotted graph metrics as the SELECT list.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 262e1a8f9a
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if resolved_metric and ref_model_name is None: | ||
| for dep in resolved_metric.get_dependencies(self.graph, model_name): | ||
| collect_measures_from_metric(dep, visited) | ||
| if resolved_metric.sql and sql_has_aggregate(resolved_metric.sql, self.dialect): | ||
| collect_sql_columns_for_model(resolved_metric.sql) | ||
| return |
There was a problem hiding this comment.
When a top-level metric resolves through the new exact graph-metric path and is a simple aggregation (for example name: finance.revenue, agg: sum, sql: orders.amount), this branch only collects dependency metrics or SQL that still contains an aggregate. The metric SQL later renders as SUM(orders_cte.amount), but _build_model_cte never projects amount, so dotted top-level aggregation metrics that this change now documents/supports generate SQL referencing a missing CTE column. Collect columns from resolved_metric.sql for resolved_metric.agg as well.
Useful? React with 👍 / 👎.
nicosuave
force-pushed
the
review/python-native-contract
branch
from
b3ee67c to
1b91738
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1b91738452
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| for model in graph.models.values(): | ||
| if any(metric.extends for metric in model.metrics): | ||
| resolved_metrics = resolve_metric_inheritance({metric.name: metric for metric in model.metrics}) |
There was a problem hiding this comment.
When load_from_directory parses native files one at a time, a child model whose extends parent is in a different file reaches this loop before model inheritance has been merged. If that child also has a metric extending a parent metric, resolve_metric_inheritance({metric.name: metric for metric in model.metrics}) only sees the child file's local metrics and raises Metric '<parent>' not found, even though the parent model will be loaded later. This breaks cross-file native model inheritance that otherwise works; skip model-local metric inheritance while a model parent is unresolved, or run it after the directory-level merge.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 62ae7e36f3
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| for metric_def in data.get("metrics") or []: | ||
| metric = self._parse_metric(metric_def) | ||
| metric = self._parse_metric(metric_def, source_path=source_path, context="metric") |
There was a problem hiding this comment.
This top-level native metrics parsing only works when SidemanticAdapter is invoked directly. In the CLI-first load_from_directory path, YAML detection still selects SidemanticAdapter only when a file has top-level models; a native file that contains only version: 1 plus graph-level metrics is instead handed to MetricFlowAdapter (sidemantic/loaders.py lines 130-134) and fails to load. This breaks valid split native projects where models and graph metrics live in separate files; detect native version/native metric shape before the MetricFlow fallback.
Useful? React with 👍 / 👎.
1 participant
Base: #183