I have list derived from JSON as below :

list = ['select', 'name1 = a.column1', 'name2 = a.column2', 'name3 = a.[column3]',
        'from', 'xyz.[Table1$Name] c',
        'select', 'name2 = b.othercolumn1', 'name2 = b.[othercolumn2]', 'name3 = b.othercolumn3',
        'from', 'abc.[Table2$Name] d',
        'where', 'x.[TableX] = '123',     and so on.....]

The output I want is a data frame with 2 columns which I can extract as excel as below:

Table Name         Column Name
Table1$Name        column1
Table1$Name        column2
Table1$Name        column3
Table2$Name        othercolumn1
Table2$Name        othercolumn2
Table2$Name        othercolumn3

I have tried in many ways but I am unable to achieve the desired output, I only want to have excel file output for all columns from this ".bim" file against the respective tables, everything else like where statement, [], c, a., b., should be removed in final output.

CodePudding user response：

• as per comments, you are really asking how to parse SQL into it's components
• this approach uses lark to parse. This is a sophisticated topic, so have demonstrated how to integrate with pandas in another way
• your sample SQL is not valid based on grammar I have used. IMHO this is a decent grammar for SQL. I have worked with many database engines, not one that I have worked with allows assignment in select clause or used of $ in column or table names

from lark import Lark, Tree, Visitor, Tree, Token, UnexpectedInput
import requests
import pandas as pd

grammar = requests.get(
    "https://raw.githubusercontent.com/zbrookle/sql_to_ibis/main/sql_to_ibis/grammar/sql.lark"
).text
parser = Lark(grammar, start="query_expr")


class sqlparts(Visitor):
    __columns = []
    __tables = []

    def __init__(self):
        self.__columns = []
        self.__tables = []

    def column_name(self, tree):
        self.__columns  = [
            tok.value
            for tok in tree.scan_values(
                lambda v: isinstance(v, Token) and v.type == "CNAME"
            )
        ]

    def table(self, tree):
        self.__tables  = [
            tok.value
            for tok in tree.scan_values(
                lambda v: isinstance(v, Token) and v.type == "CNAME"
            )
        ]

    def data(self):
        return {"columns": self.__columns, "tables": self.__tables}


df = pd.DataFrame(
    {
        "sql": [
            "select col1, col2, col3 from table1, table2 where col7=8",
            'select "hello" from a_long_table where col7=8',
            'select "hello" from a_long_table where col7=8 groupby col8',
            "select col$ from tables",
        ]
    }
)


def applyparse(sql):
    d = sqlparts()
    try:
        t = parser.parse(sql, start="query_expr")
        d.visit(t)
        return d.data()
    except UnexpectedInput as e:
        return {"error": e}


df.join(df["sql"].apply(applyparse).apply(pd.Series))