Fix Register converter for DataType.BIT

[bgunebakan]

Summary of the changes / Why this is an improvement

DataType.BIT had an enum member but no entry in _DEFAULT_CONVERTERS, so BIT values fell through _to_default and reached Python in CrateDB's SQL literal form, B'0110'. That form is not accepted back as a query parameter, which made the round-trip asymmetric.

cur.execute("SELECT flags FROM t")            # -> "B'00000001'"
cur.execute("... WHERE flags = ?", (value,))  # ProgrammingError: Cannot cast value
_to_bit_string now strips the literal wrapper and returns a plain string of 0/1 digits

_to_bit_string now strips the literal wrapper and returns a plain string of 0/1 digits, so what you read is what you can write.

Checklist

• Link to issue this PR refers to (if applicable): Fixes Register converter for DataType.BIT #817
• Increased test coverage
• Added required documentation

[mfussenegger]

Left two suggestions - otherwise lgtm

[mfussenegger]

This looks rather odd as documentation - especially the cursor.execute('') will likely be rather confusing. Can we hide the set_next_response and have a cursor.execute("select b'0110'") instead?

[bgunebakan]

Thanks, you are right. I changed it to cursor.execute("select b'0110'"). I didn't understand what you meant "hide the set_next_response"

[mfussenegger]

hide: directives like we use in the CrateDB docs. E.g.:

https://github.com/crate/crate/blob/8900fd40976c994d14a2097daff59ea32af3bff8/docs/general/information-schema.rst?plain=1#L50-L50

This way it won't show up in the rendered documentation.

[mfussenegger]

Suggested change

	_BIT_LITERAL = re.compile(r"B'([01]*)'")
	_BIT_LITERAL = re.compile(r"B'([01]+)'")

I think empty bitstrings are not possible? But please double check

[florinutz]

hey, I checked, they work:

>>> SELECT B''
{"cols":["B''"],"col_types":[25],"rows":[["B''"]],"rowcount":1}

>>> SELECT ''::bit(0)
{"cols":["CAST('' AS bit(0))"],"col_types":[25],"rows":[["B''"]],"rowcount":1}

>>> SELECT B'0110'::bit(0)
{"cols":["CAST(B'0110' AS bit(0))"],"col_types":[25],"rows":[["B''"]],"rowcount":1}

>>> CREATE TABLE tabel (b bit(0))
{"cols":[],"col_types":[],"rows":[[]],"rowcount":1}

---

and unrelated to this ticket:

>>> SELECT ''::bit(1)
{"error":{"message":"StringIndexOutOfBoundsException[Index 0 out of bounds for length 0]","code":5000}}

>>> SELECT NULL::bit(4)
{"error":{"message":"NullPointerException[Cannot invoke \"io.crate.sql.tree.BitString.length()\" because \"bs\" is null]","code":5000}}

[mfussenegger]

Both will be fixed in 6.4.5 with crate/crate#20155 and crate/crate#20156

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request

[florinutz]

looks good!

the 2 statements I found return http 500 with a java exception instead of a SQL error (crate 6.4.1). They are out of the scope of this ticket, and I think we should raise it as bug(s?) for crate/crate

[bgunebakan]

looks good!

the 2 statements I found return http 500 with a java exception instead of a SQL error (crate 6.4.1). They are out of the scope of this ticket, and I think we should raise it as bug(s?) for crate/crate

I also reproduced, and yes, looks like a bug. I think crate's 500 codes shouldn't reach all the way here, they should report as a 4000 error. Could you create a bug report on crate/crate?