fix(cubesql): Support EXTRACT(EPOCH) over timestamp diff on Snowflake

[MazterQyou]

Check List

• Tests have been run in packages where changes have been made if available
• Linter has been run for changed code
• Tests for the changes have been added if not covered yet
• Docs have been added / updated if required

Description of Changes Made

This PR detects EXTRACT(EPOCH FROM (a - b)) over a timestamp subtraction in the wrapper and, when the dialect provides an expressions/extract_epoch_diff template, renders the operands separately as a seconds diff. This is needed because Snowflake accepts EPOCH on a timestamp but not on an interval. Snowflake maps it to TIMESTAMPDIFF(MICROSECOND, b, a) / 1000000; all other dialects fall back to the existing EXTRACT(EPOCH FROM (a - b)) output unchanged. Related test is included.

[claude]

Claude finished @MazterQyou's task in 50s —— View job

---

Code Review

Verdict: LGTM — 0 high, 0 medium, 2 low (nits)

• Read PR diff
• Inspect surrounding wrapper logic
• Verify template-existence pattern matches existing usage
• Check edge cases (Cast peeling, non-diff EPOCH, SQL injection)

Summary

Clean, well-scoped change. The detection path is correctly gated: dialect must opt in via expressions/extract_epoch_diff, otherwise behavior is unchanged. The SQL-injection guard on date_part runs before the new branch (wrapper.rs:3139), so the new code only executes for the literal "epoch". args[1] is safe — guarded by args.len() >= 2. The contains_template opt-in pattern is consistent with the existing expressions/interval handling in transport/service.rs:802.

The timestamp_diff_operands helper recursively peels Cast wrappers, which correctly handles EXTRACT(EPOCH FROM CAST(a - b AS …)) and falls through cleanly for non-subtraction inputs (regular EXTRACT(EPOCH FROM ts_column) still uses the standard expressions/extract template). Good fallback hygiene.

Snowflake template DATEDIFF(SECOND, {{ right }}, {{ left }}) correctly inverts operand order to match left - right semantics.

The test covers both arms — generic dialect keeps EXTRACT(epoch …), Snowflake-style dialect emits DATEDIFF(SECOND, …).

Nits (low)

1. wrapper.rs:3066-3078 — timestamp_diff_operands only matches BinaryExpr with Operator::Minus. Semantically equivalent forms like Expr::Negative + addition, or DataFusion's date_sub/timestamp_sub scalar functions, won't be routed through the new path. Probably fine since the rewriter normalizes the typical incoming shape, but worth noting if Snowflake users report EXTRACT(EPOCH FROM <interval>) slipping through from other expression shapes.

2. wrapper.rs:3144-3151 — Minor: sql_generator.get_sql_templates() is called twice (once for contains_template, once for extract_epoch_diff_expr). Not hot-path, just slightly noisy. A local binding would tidy it up.
   • Branch: cubesql/snowflake-extract-epoch-diff

[codecov]

Codecov Report

❌ Patch coverage is 86.74699% with 11 lines in your changes missing coverage. Please review.
✅ Project coverage is 83.67%. Comparing base (28c6938) to head (d14c442).
⚠️ Report is 25 commits behind head on master.

Files with missing lines	Patch %	Lines
...t/cubesql/cubesql/src/compile/engine/df/wrapper.rs	73.68%	10 Missing ⚠️
rust/cubesql/cubesql/src/compile/mod.rs	97.22%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #11105      +/-   ##
==========================================
+ Coverage   83.61%   83.67%   +0.06%     
==========================================
  Files         256      257       +1     
  Lines       78904    79360     +456     
==========================================
+ Hits        65975    66405     +430     
- Misses      12929    12955      +26     

Flag	Coverage Δ
cubesql	83.67% <86.74%> (+0.06%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[claude]

Claude finished @MazterQyou's task in 56s —— View job

---

Code Review

Verdict: LGTM — 0 high, 0 medium, 2 low (nits). Re-affirmed on PR sync; head commit unchanged since prior review.

Details

• Read PR diff
• Inspect surrounding wrapper logic
• Verify template-existence pattern matches existing usage
• Check edge cases (Cast peeling, non-diff EPOCH, SQL injection)
• Re-verify against current HEAD (465ba42 — same single commit)

Summary

Clean, well-scoped change. The detection path is correctly gated: dialect must opt in via expressions/extract_epoch_diff, otherwise behavior is unchanged. The SQL-injection guard on date_part runs before the new branch (wrapper.rs:3137), so the new code only executes for DATE_PART_REGEX-validated parts (and matches "epoch" case-insensitively). args[1] is safe — guarded by args.len() >= 2. The contains_template opt-in pattern is consistent with the existing expressions/interval handling in transport/service.rs.

timestamp_diff_operands recursively peels Cast wrappers, correctly handling EXTRACT(EPOCH FROM CAST(a - b AS …)) and falling through cleanly for non-subtraction inputs (regular EXTRACT(EPOCH FROM ts_column) keeps the standard expressions/extract template). Good fallback hygiene.

Snowflake template DATEDIFF(SECOND, {{ right }}, {{ left }}) correctly inverts operand order to match left - right semantics.

The test covers both arms — generic dialect keeps EXTRACT(epoch …), Snowflake-style dialect emits DATEDIFF(SECOND, …).

Nits (low)

1. wrapper.rs:3070-3080 — timestamp_diff_operands only matches BinaryExpr with Operator::Minus (plus Cast peel). Semantically equivalent forms — Expr::Negative + addition, or DataFusion date_sub/timestamp_sub scalar functions — won't route through the new path. Probably fine since the rewriter normalizes the typical incoming shape, but worth noting if Snowflake users report EXTRACT(EPOCH FROM <interval>) still slipping through.

2. wrapper.rs:3148-3168 — Minor: sql_generator.get_sql_templates() is called twice (once for contains_template, once for extract_epoch_diff_expr). Not hot-path, just slightly noisy. A let templates = sql_generator.get_sql_templates(); local binding would tidy it up.

• Branch: `cubesql/snowflake-extract-epoch-diff`

[claude]

Claude finished @MazterQyou's task in 29s —— View job

---

Code Review

Verdict: LGTM — 0 high, 0 medium, 2 low (nits). Re-affirmed on PR sync; head commit unchanged (d14c442) since prior review.

Details

• Read PR diff
• Inspect surrounding wrapper logic
• Verify template-existence pattern matches existing usage
• Check edge cases (Cast peeling, non-diff EPOCH, SQL injection)
• Re-verify against current HEAD (d14c442 — still the only commit on the branch)

Summary

Clean, well-scoped change. The detection path is correctly gated: dialect must opt in via expressions/extract_epoch_diff, otherwise behavior is unchanged. The SQL-injection guard on date_part runs before the new branch in wrapper.rs, so the new code only executes for DATE_PART_REGEX-validated parts (and matches "epoch" case-insensitively). args[1] is safe — guarded by args.len() >= 2. The contains_template opt-in pattern is consistent with the existing expressions/interval handling in transport/service.rs.

timestamp_diff_operands recursively peels Cast wrappers, correctly handling EXTRACT(EPOCH FROM CAST(a - b AS …)) and falling through cleanly for non-subtraction inputs (regular EXTRACT(EPOCH FROM ts_column) keeps the standard expressions/extract template). Good fallback hygiene.

Snowflake template DATEDIFF(SECOND, {{ right }}, {{ left }}) correctly inverts operand order to match left - right semantics.

The test covers both arms — generic dialect keeps EXTRACT(epoch …), Snowflake-style dialect emits DATEDIFF(SECOND, …).

Nits (low)

1. wrapper.rs timestamp_diff_operands — only matches BinaryExpr with Operator::Minus (plus Cast peel). Semantically equivalent forms — Expr::Negative + addition, or DataFusion date_sub/timestamp_sub scalar functions — won't route through the new path. Probably fine since the rewriter normalizes the typical incoming shape, but worth noting if Snowflake users report EXTRACT(EPOCH FROM <interval>) still slipping through.

2. wrapper.rs extract branch — Minor: sql_generator.get_sql_templates() is called twice (once for contains_template, once for extract_epoch_diff_expr). Not hot-path, just slightly noisy. A let templates = sql_generator.get_sql_templates(); local binding would tidy it up.

• Branch: cubesql/snowflake-extract-epoch-diff •