Limitations

Coral has the following constraints:

• Native builds. drop_column rejects under chelis build; run code that needs it with chelis eval or chelis test. Native describe examples cover float and integer columns. Multi-aggregation examples cover string and integer group keys; join examples cover string keys with partial and complete matches. CSV and JSON reader examples cover two- and three-row frames with string and float columns. These examples do not establish native support for every combination of column types and shapes.
• Concatenation. Pass frames with the same column names, types, and row count. concat matches columns by name using the first frame's columns: an extra column in a later frame is omitted without an error, while a missing column or a different type fails. See Concatenation.
• File output. The CSV and JSON writers ignore an integer column's missing-value mask, so a masked zero writes as 0 and a read does not restore the mask. The CSV writer does not quote column names. The JSON writer escapes column names and string cells, and writes NaN, inf, and -inf as null. A JSON value such as 1e39 reads as an f32 infinity, so it writes back as null. The I/O chapter gives supported inputs.
• Large files. Reading or writing large, wide frames can take substantial time.
• Column types. GroupBy cannot return bool group keys. Bool non-key columns cannot pass through joins; inner and left joins also reject bool keys. pivot and melt take float value columns and string id/index columns, so stack needs every column to be float. String and bool columns have no missing marker.
• Missing values. Float NaNs are not skipped by GroupBy aggregations or by the rolling functions; masked integers are skipped by GroupBy. Missing keys group together and match each other in joins. See Missing values.
• Unchecked inputs. Coral does not check that Coral.AsOf right-hand keys are sorted, that an ewm alpha lies in (0, 1], or that a with_column column has the frame's row count. A wrong input gives a wrong result, not an error. The file writers are the exception: write_csv_frame and write_json_frame refuse a frame whose columns have different lengths instead of writing a file truncated to the shortest one.
• Output conventions. outer_join returns its key as a string column, whatever the input key type. from_columns and empty use dictionary entry order for columns; use from_pairs for explicit order. In a multi-aggregation, each value column can appear only once, and an AggCount spec needs a numeric value column. String sorting uses Coral's own character ranking; see Filtering.