Parser: Type inference is locale-dependent and mishandles edge cases (inf, nan, leading spaces)

[im-anishraj]

Description

The infer_type function uses std::strtoll and std::strtod which are locale-dependent on some platforms. In European locales where , is the decimal separator, "3.14" may be inferred as STRING instead of FLOAT64.

Additionally, the inference has edge cases:

• "inf" and "nan" are parsed as valid floats by strtod but should arguably be STRING
• " 123" (leading space) is treated as STRING because the space isn't trimmed before inference
• "1e10" is correctly inferred as FLOAT64 but "1E10" behavior may vary by platform

Acceptance Criteria

• Use std::strtod with "C" locale explicitly, or use a locale-independent parser
• Trim whitespace from values before type inference
• "inf", "-inf", "nan" are treated as FLOAT64 (matching pandas)
• Add tests for locale-edge-case values
• Add tests for scientific notation variants

Difficulty: Intermediate
Files: cpp/src/csv_reader.cpp, tests/test_csv.py

[KamnaJ5]

Hi @im-anishraj, I’d like to work on this issue.

I reviewed the parser edge cases mentioned in the description, including locale-dependent float parsing, whitespace handling, and scientific notation behavior. I plan to investigate the current inference logic and add tests for the identified edge cases to ensure consistent parsing behavior across platforms.

Please assign this issue to me. Thank you

[im-anishraj]

Thanks @KamnaJ5.

This is a parser-level bug and needs a focused C++ plan before assignment. Please choose one issue first; if this is the one, include the parser files, locale/inf/nan behavior, whitespace handling, and regression tests you would add.