By Re'em Kishnevsky And Maayan Kestenberg

The table file format is as follows:

• Each table directory contains a .col file for each column, and an additional .pointers file for each VARCHAR column.

  • .col Files:
    .col files are merely the entire column values dumped into a file, one after the other. If the column is of a fixed-length type (INT | FLOAT | TIMESTAMP) then the offsets can be calculated by the formula: i*64. (Where i is the index of the desired value). If the column is of a variable-length type (VARCHAR) then the offsets cannot be calculated and thus stored in a .pointers file. INT, FLOAT and TIMESTAMP columns are stored as a .scol file (Static Column).
    .col file format:

    [record(0)][record(1)]...[record(N-1)]
    

    N = Number of records in the table
    record(X) = value of the X-th record in the column

  • .pointers Files:
    .pointers files contain a sequence of pointers (addresses / offsets). each pointer is a 64 bit unsigned int that is an address of a record value in the .col file that coresponds to the column - allowing for a maximum of about 18 quintillion varchar records in the column. In this way, the address of the i-th record of the column is the value of the pointer at the address i*64.
    .pointers File Format:

    [pointer(1)][pointer(2)]...[pointer(N-1)]
    

    N = Number of records in the table
    pointerX = 64 bit unsigned int address of the X-th record in the column
    (first pointer points to record(1) since record(0) is always at offset 0).

Currenly, the project's features are:

• Command Line Interface with arguments -v, -r, -d, -h.
• SQL Commands: CREATE, CREATE AS SELECT, LOAD, DROP.
• Select command only supports selecting all the columns (*) and the clauses: INTO OUTFILE, WHERE.
• Pretty print of the select output to the terminal (Works better on Unix).