fix: include empty or null values

[ElfenB]

fixes #4123

[vercel]

@ElfenB is attempting to deploy a commit to the Umami Software Team on Vercel.

A member of the Team first needs to authorize it.

[greptile-apps]

Greptile Summary

This PR fixes filter queries in both the ClickHouse and Prisma (PostgreSQL) database layers so that rows with absent/empty values are correctly included when applying negative filter operators (notEquals, doesNotContain).

Key changes:

• prisma.ts: Adds OR column IS NULL to both notEquals and doesNotContain clauses. This is a necessary fix because PostgreSQL's three-valued logic causes NULL != value and NULL NOT ILIKE value to evaluate to NULL (falsy), silently dropping rows with NULL column values from negative-filter results.
• clickhouse.ts: Adds OR column = '' to the notEquals clause only. ClickHouse website_event string columns are declared as non-nullable (String, not Nullable(String)), so '' is the correct "no value" sentinel rather than NULL. The doesNotContain case intentionally requires no fix: positionCaseInsensitive('', value) already returns 0, so empty-string rows naturally pass the = 0 check.

Minor concern: In ClickHouse, if a user ever provides an empty string '' as the filter value for notEquals, the generated SQL (column != '' OR column = '') is a tautology that matches every row. This is a pre-existing edge case amplified slightly by the fix, but unlikely to surface in practice given how filters are populated from UI inputs.

Confidence Score: 5/5

Safe to merge — both fixes are logically correct for their respective databases and address the reported bug.

The Prisma fix is straightforward and correct: adding OR IS NULL is the standard SQL pattern for including NULL rows in negative comparisons. The ClickHouse fix correctly uses OR = '' to match the non-nullable String column design. The only remaining concern (tautology when filtering by empty string in ClickHouse) is a P2 edge case that does not affect normal usage.

No files require special attention, though src/lib/clickhouse.ts has a minor edge-case noted above.

Important Files Changed

Filename	Overview
src/lib/clickhouse.ts	Fixed notEquals to include empty-string rows (OR column = ''). The doesNotContain case needs no fix because positionCaseInsensitive('', value) = 0 is already TRUE for empty strings. Minor edge case: filtering "not equals empty string" now produces a tautology that matches all rows.
src/lib/prisma.ts	Fixed notEquals and doesNotContain to include NULL rows via OR column IS NULL. Correct and necessary: in SQL, NULL != value and NULL NOT ILIKE value both evaluate to NULL (falsy), silently excluding NULL rows from these negative filters.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[mapFilter called] --> B{operator}
    B --> |equals| C["column = value"]
    B --> |notEquals - Prisma| D["(column != value OR column IS NULL)"]
    B --> |notEquals - ClickHouse| E["(column != value OR column = '')"]
    B --> |contains| F["column ilike value / positionCaseInsensitive > 0"]
    B --> |doesNotContain - Prisma| G["(column NOT ilike value OR column IS NULL)"]
    B --> |doesNotContain - ClickHouse| H["positionCaseInsensitive = 0"]

    note1["ClickHouse String columns are non-nullable\n→ empty string '' is the 'no value' sentinel\n→ positionCaseInsensitive('', v) = 0 already TRUE\n   so doesNotContain needs no fix"]
    note2["Postgres columns are nullable\n→ NULL rows silently dropped by != and NOT ILIKE\n→ OR IS NULL fix required for both operators"]

    E -.-> note1
    D -.-> note2
    G -.-> note2

Loading

_{Reviews (1): Last reviewed commit: "fix: include empty or null values" | Re-trigger Greptile}

[greptile-apps]

Tautology when filtering by empty string

The guard OR ${column} = '' is the ClickHouse equivalent of IS NULL for Postgres, but it introduces a tautology when the filter value itself is an empty string ''. In that case the generated SQL becomes:

(column != '' OR column = '')

which is always TRUE for every row — so a "not equals empty-string" filter would return the entire dataset instead of excluding nothing. In practice users are unlikely to filter on an exact empty-string value, but it is worth being aware of this edge case. The Prisma fix avoids this because IS NULL only catches NULL, not the '' value itself.