[#29816]: Add CAST to list of supported ClickHouse functions

[matt-metivier]

Problem

Users cannot use CAST(column AS Type) syntax in HogQL. The immediate need is casting String columns to JSON type for the Query API. We intentionally avoid exposing raw CAST syntax to keep type strings non-user-controlled.

Flow:

1. User writes HogQL: toJSON(properties)
2. Parser: Creates ast.Call(name="toJSON", args=[properties])
3. Function Registry: conversions.py registers "toJSON": HogQLFunctionMeta("toJSON", 1, 1)
4. Printer (base.py line 523): Special handling converts to CAST({expr} AS JSON)
5. Output: CAST(events.properties AS JSON)

Closes #29816

Changes

• Added toJSON(x) HogQL function that prints CAST(x AS JSON) in ClickHouse
• Kept CAST(expr AS type) syntax unsupported to avoid user-controlled type strings
• Added parser test asserting CAST syntax is rejected
• Added printer test asserting toJSON() outputs CAST(... AS JSON)

How did you test this code?

Added unit tests in:

• posthog/hogql/test/_test_parser.py - parser test for CAST rejection
• posthog/hogql/printer/test/test_printer.py - printer test for toJSON() output

Publish to changelog?

Yes

Notes

• This is a safe workaround for the issue: users can use toJSON(properties) instead of CAST(properties AS JSON).
• Docs update needed in posthog.com repo: contents/docs/sql/clickhouse-functions.mdx -https://github.com/PostHog/posthog.com/pull/14549/changes

[greptile-apps]

_{3 files reviewed, 3 comments}

_{Edit Code Review Agent Settings | Greptile}

[matt-metivier]

@sakce or @orian Whenever you've time, can you give me a review. First time doing any work on PostHog, please let me know if their is anything you want different to match posthog style.

I did review this before writing the pr- https://posthog.com/handbook/engineering/conventions/backend-coding

[rafaeelaudibert]

Pinginfg @Gilbert09 since I considered doing this in the past and was told this was possibly too dangerous

[matt-metivier]

Pinginfg @Gilbert09 since I considered doing this in the past and was told this was possibly too dangerous

I was taking another look through the code, I do see y'all hardcode in conversions.py to make sure the type string is never user-controlled.

• https://github.com/PostHog/posthog/blob/master/posthog/hogql/functions/clickhouse/conversions.py#L40

I guess the risk with CAST(expr AS Type) is that the type is a raw string that gets inserted into the SQL. Even with validation, there's that risk.

[matt-metivier]

Going to update the code to instead of CAST(x AS JSON), add explicit functions:

"castToJSON": HogQLFunctionMeta("CAST", 1, 1, suffix_args=[ast.Constant(value="JSON")]),
"castToString": HogQLFunctionMeta("CAST", 1, 1, suffix_args=[ast.Constant(value="String")]),

Then a user would write: castToJSON(properties) instead of CAST(properties AS JSON)

[georgemunyoro]

@matt-metivier Appreciate the contribution! I don’t have full ClickHouse context here (Tom will have more insight), but language-wise you’ll also need to update the c++ parser. While we keep the parsers in sync, in practice we currently really only use the c++ one.

That said, we’re in the middle of transitioning to a new, language-agnostic parser that outputs JSON instead of python objects. Because of that, there’s no need to update the legacy c++ parser, just the new JSON one.

[matt-metivier]

• contents/docs/sql/clickhouse-functions.mdx

Didn't realize this, thank you for mentioning it. I updated it and I opened a pr to update the docs in case we decide to go with this path pending @Gilbert09 approval.
PostHog/posthog.com#14549

Flow:

1. User writes HogQL: toJSON(properties)
2. Parser: Creates ast.Call(name="toJSON", args=[properties])
3. Function Registry: conversions.py registers "toJSON": HogQLFunctionMeta("toJSON", 1, 1)
4. Printer (base.py line 523): Special handling converts to CAST({expr} AS JSON)
5. Output: CAST(events.properties AS JSON)