Question

Allow my to elaborate on the question with an example. I am writing from the
perspective that

(1) bytecodes should not be used to implement the logic already
implemented (hopefully more efficiently) in the database engine (e.g., if we
need to filter out 20 lines out of 500 coming out as the result of an SQL query,
we should be writing a better where clause), and that

(2) I have only a conceptual
understanding of foreign keys (e.g., they automatically create, manage and enforce
the constraints required to maintain data integrity across different tables).

Now, lets consider a simple schema with 4 tables and 15 columns in them
as follows (assume all columns are not null):

people
    pid bigint autoinc PK
    fname   varchar(32)
    lname   varchar(32)
    dob date
    sex char(1)
    addyID  bigint FK to address(aid)

relationship
    relof   bigint FK to people(pid)
    relto   bigint FK to people(pid)
    relis   tinyint

address
    aid bigint autoinc PK
    street  varchar(128)
    zipcode int FK to z2cs(zip)

z2cs
    zip int PK
    city    varchar(64)
    state   char(2)

Expectations (Please answer True/False to the following 6)

1. It is possible to create the table relationship as above with foreign keys
   pointing to different rows of the same table
2. Adding a row to relationship throws SQLException if either relof or relto
   is missing in people
3. If we add a row to address, we do not need to check if the zipcode is there in z2cs –
   missing zip throws an SQLException
4. Attempt to delete a row from address throws an exception if aid is used in people
5. Attempt to delete a row from people causes error if pid is used in relationship in
   either of the columns relto or relof
6. An SQL statement exists that lets you find which columns in a table are foreign keys
   and which table(column) do they refer to (I know “describe tablename” doesn’t do it,
   although I feel that it should)

If the answer to any of the above is false, can you please guide me to finding
what exactly is the contract for foreign keys in MySql. Thank you.

PS: If you answer True to [6], would you please share what is that statement.

Bonus question (unrelated to FK):

How would you write the create statement for the relationship table that satisfies the following constraints:

1. relof and relto cannot be the same
2. Combination of relof and relto is unique, i.e.,
   if their is a row with [relof=584,relto=7823], you cannot insert another row with
   either [relof=584,relto=7823] or [relof=7823,relto=584]

Answer 1

1. It is possible to create the table relationship as above with foreign keys pointing to different rows of the same table

NO, a FK can only point to one row. But a row can be referenced by many other rows, like, an adress can be referenced by many people.

2. Adding a row to relationship throws SQLException if either relof or relto is missing in people

YES.

3. If we add a row to address, we do not need to check if the zipcode is there in z2cs – missing zip throws an SQLException

YES.

4. Attempt to delete a row from address throws an exception if aid is used in people

YES, if there is no Cascade Delete in the FK, which then automatically would delete the
entry in people instead of failing.

5. Attempt to delete a row from people causes error if pid is used in relationship in either of the columns relto or relof

YES, see 4.

6. An SQL statement exists that lets you find which columns in a table are foreign keys and which table(column) do they refer to (I know “describe tablename” doesn’t do it, although I feel that it should)

AFAIK. There should be an information schema in MySQL where you can query the information from.

B1. relof and relto cannot be the same

Use a Trigger on update or insert to validate this. PostgreSQL knows CHECK constraints, which might ba also available in MySQL, but else a trigger is your choice.

B2. Combination of relof and relto is unique

A combined UNIQUE INDEX on both columns is your friend here.