Dr Paolo Guagliardo. Fall 2018

Transcription

1 The NULL vale Dr Paolo Gagliardo Fall 2018 NULL: all-prpose marker to represent incomplete information Main sorce of problems and inconsistencies... this topic cannot be described in a manner that is simltaneosly both comprehensive and comprehensible Those SQL featres are... fndamentally at odds with the way the world behaves C. Date & H. Darwen, Gide to SQL Standard

2 What does NULL mean? Depending on the context: Missing vale there is a vale, bt it is crrently nknown Non-applicable there is no vale (ndefined) Bt it also behaves as: Constant like any other vale Unknown a trth-vale in addition to Tre and False Meta-incompleteness We never really know what NULL means becase the meaning is ltimately defined by the application Bt we mst know how NULL behaves according to the Standard and this behavior depends on the context in which it is sed Missing vale vs. Non-applicable Person ID Name Phone 1 Jane NULL Does Jane have a phone? 2 John NULL Does John have a phone? There is no way of knowing whether a NULL here means that there is a crrently nknown vale for phone (missing vale) or there is no vale for phone (non-applicable vale)

3 NULL and schema design Person ID Name HasPhone Phone 1 Jane Yes NULL missing 2 John No NULL non-applicable What if we want Phone to be NOT NULL when HasPhone is Yes? we cannot jst declare Phone as NOT NULL we need to se a CHECK constraint We also want to check that Phone is NULL when HasPhone is No NULL and schema design Person ID Name HasPhone Phone 1 Jane NULL NULL 2 John NULL We don t know whether John and Jane have a Phone NULL in colmn HasPhone represents a missing vale What does NULL in colmn Phone mean? What abot the vale for John s Phone? Rle ot these cases with a CHECK constraint Declare HasPhone to be NOT NULL (we cannot say we don t know whether a person has a Phone)

4 NULL and schema design Getting rid of non-applicable vales Person PersonWithPhone ID Name ID Name Phone 2 John 1 Jane NULL Pros: NULLs in Phone represent missing vales Phone can be declared NOT NULL if needs be Cons: Need to make sre there is no overlap (assertion? trigger?) How do we say I don t know whether John has a Phone? (we cold add a colmn HasPhone to Person...) NULL and schema design Getting rid of non-applicable vales Person PersonWithPhone ID Name ID Phone 1 Jane 2 John 1 NULL PersonWithPhone(ID) REFERENCES Person(ID) Pros: NULLs in Phone represent missing vales Phone can be declared NOT NULL if needs be Cons: How do we say I don t know whether John has a Phone? (we cold add a colmn HasPhone to Person...)

5 Limitations of SQL s NULL as missing vales Name Person ge Jane NULL John NULL Mary 27 Carl NULL ge of Jane, John and Carl is nknown We know Jane and John have the same age Marked nlls (not part of SQL) Each missing vale has an identifier llow cross-referencing of missing vales NULL and constraints Nlls are not allowed in primary keys CRETE TBLE R ( INT PRIMRY KEY ); INSERT INTO R VLUES (NULL); ERROR: nll vale in colmn "a" violates not-nll constraint Nlls seem to behave as (distinct) missing vales with UNIQUE CRETE TBLE R ( INT UNIQUE ); INSERT INTO R VLUES (NULL); INSERT INTO R VLUES (NULL); R NULL NULL bt in fact this is simply becase NULLs are ignored

6 NULL and constraints R B 1 1 S B 1 NULL 2 NULL S(,B) REFERENCES R(,B) Is NULL treated as a missing vale here? Not really! The above instance is legal w.r.t. the FK constraint NULL and arithmetic operations Every arithmetic operation that involves a NULL reslts in NULL SELECT 1+NULL S sm, 1-NULL S diff, 1*NULL S mlt, 1/NULL S div sm diff mlt div NULL NULL NULL NULL (1 row) Observe that SELECT NULL/0 also retrns NULL instead of throwing a DIVISION BY ZERO error! Here, NULL is treated as an ndefined vale

7 NULL and aggregation (1) ggregate fnctions ignore nlls Consider R = {0, NULL, 1, NULL} on attribte SELECT MIN(), MX(), COUNT(), SUM(), CST( VG() S nmeric(2,1) ) FROM R ; min max cont sm avg (1 row) NULL and aggregation (2) ggregate fnctions ignore nlls Consider R = {0, NULL, 1, NULL} on attribte Exception: SELECT COUNT(*) FROM R ; cont (1 row)

8 NULL and aggregation ggregation (except COUNT) on an empty bag reslts in NULL Consider R = {0, 1, NULL} on attribte SELECT MIN(), MX(), SUM(), VG(), COUNT() FROM R WHERE = 2 ; min max sm avg cont NULL NULL NULL NULL 0 (1 row) The semantics of these nlls is that of ndefined vales NULL and set operations What is the answer to Q 1 : SELECT * FROM R UNION SELECT * FROM S Q 2 : SELECT * FROM R INTERSECT SELECT * FROM S Q 3 : SELECT * FROM R EXCEPT SELECT * FROM S when R = {1, NULL, NULL} and S = {NULL}? nswer to Q 1 : {1, NULL} nswer to Q 2 : {NULL} nswer to Q 3 : {1} In set operations NULL is treated like any other vale

9 NULL and set operations What is the answer to Q 1 : SELECT * FROM R UNION LL SELECT * FROM S Q 2 : SELECT * FROM R INTERSECT LL SELECT * FROM S Q 3 : SELECT * FROM R EXCEPT LL SELECT * FROM S when R = {1, NULL, NULL} and S = {NULL}? nswer to Q 1 : {1, NULL, NULL, NULL} nswer to Q 2 : {NULL} nswer to Q 3 : {1, NULL} NULL in selection conditions (1) What is the answer to Q 1 : SELECT * FROM R, S WHERE R. = S. Q 2 : SELECT * FROM R, S WHERE R. <> S. Q 3 : SELECT * FROM R, S WHERE R. = S. OR R. <> S. when R = {1, NULL} and S = {NULL}? R. 1 NULL S. NULL = R. S. 1 NULL NULL NULL nswer to all three qeries: {}

10 NULL and comparisons SELECT 1=NULL S reslt; reslt NULL (1 row) This is not an ndefined vale it is a trth-vale: nknown SELECT 1=NULL OR TRUE S reslt; reslt t (1 row) Try: SELECT NULL/1 OR TRUE S reslt; Evalation of selection conditions SQL ses three trth vales: tre (t), false (f), nknown () 1. Every comparison (except IS [NOT] NULL and EXISTS) where one of the argments is NULL evalates to nknown 2. The trth vales assigned to each comparison are propagated sing the following tables: ND t f t t f f f f f f OR t f t t t t f t f t t f NOT f t 3. The rows for which the condition evalates to tre are retrned

11 NULL in selection conditions (2) What is the answer to Q 1 : SELECT * FROM R, S WHERE R. = S. Q 2 : SELECT * FROM R, S WHERE R. <> S. Q 3 : SELECT * FROM R, S WHERE R. = S. OR R. <> S. when R = {1, NULL} and S = {NULL}? R. S. θ 1 θ 2 θ 3 1 NULL NULL NULL NULL and qery eqivalence (1) Q 1 SELECT R. FROM R INTERSECT SELECT S. FROM S Q 2 SELECT DISTINCT R. FROM R, S WHERE R. = S. On databases withot nlls, Q 1 and Q 2 give the same answers On databases with nlls, they do not For example, when R = S = {NULL} Q 1 retrns {NULL} Q 2 retrns {}

12 NULL and qery eqivalence (2) Consider R = {1, NULL} and S = {NULL} Q 1 : Q 2 : Q 3 : SELECT R. FROM R EXCEPT SELECT S. FROM S ; SELECT DISTINCT R. FROM R WHERE NOT EXISTS ( SELECT * FROM S WHERE S.=R. ); SELECT DISTINCT R. FROM R WHERE R. NOT IN ( SELECT S. FROM S ); nswer: { 1 } nswer: { 1, NULL } nswer: {} Inner joins R B C S D SELECT * FROM R [INNER] JOIN S ON R.B = S.C ; B C D

13 Oter joins (1) R B C S D SELECT * FROM R LEFT [OUTER] JOIN S ON R.B = S.C ; B C D Same as SELECT * FROM R JOIN S ON R.B = S.C UNION LL SELECT R.*, NULL, NULL FROM R WHERE NOT EXISTS ( SELECT * FROM S WHERE R.B = S.C ) Oter joins (2) R B C S D SELECT * FROM R RIGHT [OUTER] JOIN S ON R.B = S.C ; B C D Same as SELECT * FROM R JOIN S ON R.B = S.C UNION LL SELECT NULL, NULL, S.* FROM S WHERE NOT EXISTS ( SELECT * FROM R WHERE R.B = S.C )

14 Oter joins (3) R B C S D SELECT * FROM R FULL [OUTER] JOIN S ON R.B = S.C ; B C D Same as SELECT * FROM R JOIN S ON R.B = S.C UNION LL SELECT R.*, NULL, NULL FROM R WHERE NOT EXISTS ( SELECT * FROM S WHERE R.B = S.C ) UNION LL SELECT NULL, NULL, S.* FROM S WHERE NOT EXISTS ( SELECT * FROM R WHERE R.B = S.C ) Coalescing nll vales Syntax: COLESCE(expr1, expr2) Same as CSE WHEN expr1 IS NULL THEN expr2 ELSE expr1 END Example R: 1 3 SELECT COLESCE(R.,0) S FROM R gives 1 0 3