Copyright © C. J. Date 2005page 49 SAMPLE QUERIES : l Query A:Get S#-FROM-TO triples for suppliers who have been able to supply at least one part during at least one interval of time, where FROM and TO together designate a maximal interval during which supplier S# was in fact able to supply at least one part l Query B:Get S#-FROM-TO triples for suppliers who have been unable to supply any parts at all during at least one interval of time, where FROM and TO together designate a maximal interval during which supplier S# was in fact unable to supply any part at all  You’ve got to be joking!

Copyright © C. J. Date 2005page 50 TO SUM UP : "Temporal" constraints and queries can be expressed, but they quickly get very complicated indeed We need some carefully thought out and well-designed shorthands … …which typically don’t exist yet in today’s commercial DBMSs, of course So let’s investigate!

Copyright © C. J. Date 2005page LAYING THE FOUNDATIONS :  Time and the DB  What’s the problem?  Intervals Interval operators The EXPAND and COLLAPSE operators The PACK and UNPACK operators Relational operators

Copyright © C. J. Date 2005page 52 INTERVALS : Crucial observation: Need to deal with intervals as such (i.e., as values in their own right), instead of as pairs of FROM-TO values But what’s an interval? Consider proposition:"Supplier S1 was able to supply part P1 from day 4 to day 10" Does interval "from day 4 to day 10" include day 4? day 10? Given some interval, we sometimes want to regard specified begin and end points as included and sometimes not

Copyright © C. J. Date 2005page 53 [d04:d10]--closed-closed=d04 d05 d06 d07 d08 d09 d10 [d04:d11)--closed-open=d04 d05 d06 d07 d08 d09 d10 (d03:d10]--open-closed=d04 d05 d06 d07 d08 d09 d10 (d03:d11)--open-open=d04 d05 d06 d07 d08 d09 d10 Closed-open is convenient and most often used in practice: E.g.,split [d04:d11) immediately before, say, day 7 … result is [d04:d07) and [d07:d11) But closed-closed is most intuitive and we’ll favor it throughout this presentation

Copyright © C. J. Date 2005page 54 FULLY TEMPORALIZING SUPPLIERS AND SHIPMENTS USING INTERVALS : S_DURINGSP_DURING S#DURINGS#P#DURING S1 [d04:d10]S1P1[d04:d10] S2[d02:d04]S1P2[d05:d10] S2 [d07:d10]S1P3[d09:d10] S3[d03:d10]S1P4[d05:d10] S4[d04:d10]S1P5[d04:d10] S5[d02:d10]S1P6[d06:d10] S2P1[d02:d04] S2P1[d08:d10] S2P2[d03:d03] S2P2[d09:d10] S3P2[d08:d10] S4P2[d06:d09] S4P4[d04:d08] S4P5[d05:d10]

Copyright © C. J. Date 2005page 55 PREDICATES : l S_DURING: From the begin point of DURING to the end point of DURING inclusive (and not immediately before the begin point of DURING or immediately after the end point of DURING), supplier S# was under contract l SP_DURING: From the begin point of DURING to the end point of DURING inclusive (and not immediately before the begin point of DURING or immediately after the end point of DURING), supplier S# was able to supply part P#

Copyright © C. J. Date 2005page 56 SOME IMMEDIATE ADVANTAGES : l Constraints to prohibit FROM-TO pairs in which TO < FROM are now unnecessary ("FROM < TO" is implicit) l Primary keys {S#,DURING} (for S_DURING), {S#,P#,DURING} (for SP_DURING)—choice no longer arbitrary l Don’t need to worry about whether FROM-TO intervals in previous version of DB are open or closed wrt FROM and TO [d04:d10], [d04:d11), (d03:d10], (d03:d11) are distinct "possreps" for the very same interval* —don’t need to know which, if any, is actual physical representation * See The Third Manifesto

Copyright © C. J. Date 2005page 57 INTERVALS AREN’T NECESSARILY TEMPORAL : l Tax brackets are represented by taxable income ranges (intervals whose contained points are money values) l Machines operate within certain temperature and voltage ranges (intervals whose contained points are temperatures and voltages, respectively) l Animals vary in the range of frequencies of light and sound waves to which their eyes and ears are receptive l Various natural phenomena occur in ranges in depth of soil or sea or height above sea level l Etc., etc.

Copyright © C. J. Date 2005page 58 POINT TYPES AND INTERVAL TYPES : l Granularity of interval [d04:d10] = one day = granularity of type DATE l Assume DATE is a builtin type representing Gregorian dates: i.e., points on timeline accurate to one day (granularity thus one day by definition) l Exact type of interval value [d04:d10] = INTERVAL_DATE

Copyright © C. J. Date 2005page 59 POINT TYPES AND INTERVAL TYPES (cont.) : l INTERVAL is a type generator With associated generic interval operators (see later) and constraints l DATE is the point type of this particular interval type Determines specific set of interval values that make up this particular interval type Namely, the set of all possible intervals of the form [di:dj], where di and dj are DATE values and di < dj

Copyright © C. J. Date 2005page 60 FURTHER EXAMPLES : l INTERVAL_INTEGER Values are intervals of the form [i:j], where i and j are INTEGER values and i < j Granularity = one (unity) l INTERVAL_TIMESTAMP Values are intervals of the form [ti:tj], where ti and tj are TIMESTAMP values and ti < tj Granularity = one microsecond (assume TIMESTAMP values are accurate to the microsecond)

Copyright © C. J. Date 2005page 61 POINT TYPES : Type T can be used as a point type if all of the following are defined for it: l A total ordering (">" etc. available for any pair of values of type T) l Niladic FIRST and LAST operators l Monadic NEXT and PRIOR operators (can fail) NEXT = successor function Successor function assumed unique Thus, e.g., DATE is a valid point type

Copyright © C. J. Date 2005page 62 Informal notation:Successorof d = d+1 Predecessorof d = d-1 Successor function is what enables us to determine what points are contained in any given interval l E.g., if i = [d04:d10], contained points are exactly d04, d04+1, d04+2, …, d10 l Successor function for type DATE = "next day"

Copyright © C. J. Date 2005page 63 INTERVALS : Let T be a point type. Then an interval (value) i of type INTERVAL_T is a scalar value for which two monadic ops, BEGIN and END, and one dyadic op, , are defined, such that: BEGIN(i) and END(i) each return a value of type T BEGIN(i) < END(i) If p is a value of type T, then p  i is true if and only if BEGIN(i) < p and p < END(i) are both true Note that intervals are always nonempty

Copyright © C. J. Date 2005page 64 A MORE SEARCHING EXAMPLE : A RELATION WITH TWO INTERVAL ATTRIBUTES S_PARTS_DURING S#PARTSDURING S1[P1:P3][d01:d04]Not meant to correspond S1[P2:P4][d07:d08]in any particular way S1[P5:P6][d09:d09]to sample SP_DURING value S2[P1:P1][d08:d09] S2[P1:P2][d08:d08]Note problems: e.g., "S3 S2[P3:P4][d07:d08]was able to supply P4 on S3[P2:P4][d01:d04]days 1-4" appears twice S3[P3:P5][d01:d04] S3[P2:P4][d05:d06]Will revisit this example S3[P2:P4][d06:d09]later S4 [P3:P4][d05:d08]

Copyright © C. J. Date 2005page LAYING THE FOUNDATIONS :  Time and the DB  What’s the problem?  Intervals  Interval operators The EXPAND and COLLAPSE operators The PACK and UNPACK operators Relational operators

Copyright © C. J. Date 2005page 66 INFORMAL NOTATION : Point type T, typical value p—use p+1, p+2, p-1, p-2, etc., as shorthands with obvious meanings (a real language would provide NEXT_T / PRIOR_T ops, also FIRST_T / LAST_T) Interval type INTERVAL_T—use [p1:pn] to denote typical interval selector invocation (a real language would use more explicit syntax— e.g., INTERVAL_T ( [p1:pn] ) ) Let i be the interval [b:e]. Then: BEGIN(i) and END(i) return b and e, resp. p  i  b < p AND p < e PRE(i) returns b-1 / POST(i) returns e+1 (can fail) Let i = unit interval [p:p]; POINT FROM i returns p

Copyright © C. J. Date 2005page 67 ALLEN’S OPERATORS : l i1 = i2 b1e1 i1 b2e2 i2 l i1  i2 / i2  i1 (also i1  i2 / i2  i1) b1e1 i1 b2e2 i2

Copyright © C. J. Date 2005page 68 ALLEN’S OPERATORS (cont.) : l i1 BEFORE i2 / i2 AFTER i1 b1e1b2 e2 i1i2 l i1 MEETS i2 / i2 MEETS i1 b1e1b2e2 i1 i2 l i1 OVERLAPS i2 / i2 OVERLAPS i1 b1e1 i1 b2 e2 i2

Copyright © C. J. Date 2005page 69 ALLEN’S OPERATORS (cont.) : l i1 MERGES i2  i1 OVERLAPS i2 OR i1 MEETS i2 b1e1 i1 b2e2 i2 Or: b1e1b2e2 i1 i2

Copyright © C. J. Date 2005page 70 ALLEN’S OPERATORS (cont.) : l i1 BEGINS i2 b1e1 i1 b2e2 i2 l i1 ENDS i2 b1e1 i1 b2 e2 i2

Copyright © C. J. Date 2005page 71 OTHER OPERATORS : l COUNT(i) /* aka DURATION(i) */ returns no. of points in i (i.e., cardinality) l i1 UNION i2 returns [MIN(b1,b2):MAX(e1,e2)] if i1 MERGES i2 and is otherwise undefined /* result is an interval */ b1e1 i1 b2e2 i2 b1e2 i1 UNION i2

Copyright © C. J. Date 2005page 72 OTHER OPERATORS (cont.) : l i1 INTERSECT i2 returns [MAX(b1,b2):MIN(e1,e2)] if i1 OVERLAPS i2 and is otherwise undefined /* result is an interval */ b1e1 i1 b2e2 i2 b2e1 i1 INTERSECT i2

Copyright © C. J. Date 2005page 73 OTHER OPERATORS (cont.) : l i1 MINUS i2 returns[b1:MIN(b2-1,e1)] if b1 < b2 and e1 < e2, [MAX(e2+1,b1):e1]if b1 > b2 and e1 > e2, and is otherwise undefined (i.e., undefined if i1 BEGINS i2 or i1 ENDS i2 or if either of i1 and i2 properly includes the other) /* result is an interval */ b1e1 i1 b2e2 i2 b1 b2-1 i1 MINUS i2

Copyright © C. J. Date 2005page 74 SAMPLE QUERIES : l Get supplier numbers for suppliers who were able to supply part P2 on day 8 ( SP_DURING WHEREP# = P# (‘P2’) ANDd08  DURING ) { S# } l Get pairs of suppliers who were able to supply the same part at the same time WITHSP_DURINGRENAME ( S# AS X#, DURING AS XD ) AS T1, SP_DURINGRENAME ( S# AS Y#, DURING AS YD ) AS T2, T1 JOIN T2 AS T3, (T3 WHERE XD OVERLAPS YD ) AS T4, (T4 WHERE X# < Y# ) AS T5 : T5 { X#, Y# } Note the use of WITH to introduce names for expressions

Copyright © C. J. Date 2005page 75 l Get pairs of suppliers who were able to supply the same part at the same time, together with the parts and times in question WITHSP_DURINGRENAME ( S# AS X#, DURING AS XD ) AS T1, SP_DURINGRENAME ( S# AS Y#, DURING AS YD ) AS T2, T1 JOIN T2 AS T3, (T3 WHERE XD OVERLAPS YD ) AS T4, (T4 WHERE X# < Y# ) AS T5, ( EXTEND T5 ADD ( XD INTERSECT YD AS DURING ) ) AS T6 : T6 { X#, Y#, P#, DURING }

Copyright © C. J. Date 2005page LAYING THE FOUNDATIONS :  Time and the DB  What’s the problem?  Intervals  Interval operators  The EXPAND and COLLAPSE operators The PACK and UNPACK operators Relational operators

Copyright © C. J. Date 2005page 77 THE EXPAND AND COLLAPSE OPERATORS : Work on sets of intervals, not individual intervals (or pairs of intervals) per se Each takes a set of intervals all of the same type as its single operand and returns another such set as its result { […], […], …, […] } Result in each case is a particular canonical form for the original set

Copyright © C. J. Date 2005page 78 ASIDE : "CANONICAL FORM" Given(a) a set S of objects and (b) a notion of equivalence among such objects subset C of S is said to be a set of canonical forms for S (under the stated definition of equivalence) if and only if every object s in S is equivalent to just one object c in C l Object c is the canonical form for object s l All "interesting" properties that apply to s also apply to c; thus, we can study just the small set C, not the large set S, in order to obtain or prove a variety of "interesting" results

Copyright © C. J. Date 2005page 79 EXPANDED FORM : The objects we wish to study are sets of intervals, where the intervals are all of the same type Let X1 and X2 be two such sets. Define equivalence: X1 and X2 are equivalent if and only if set of all points in intervals in X1 = set of all points in intervals in X2 E.g.: X1={ [d01:d01], [d03:d05], [d04:d06] } X2={ [d01:d01], [d03:d04], [d05:d05], [d05:d06] }

Copyright © C. J. Date 2005page 80 EXPANDED FORM (cont.) : Corresp set of points = { d01, d03, d04, d05, d06 } But we’re more interested in corresp set of unit intervals: X3 = { [d01:d01], [d03:d03], [d04:d04], [d05:d05], [d06:d06] } X3 is equivalent to both X1 and X2 (it’s the expanded form of both) If X is a set of intervals all of the same type, then the expanded form of X is the set of all intervals of the form [p:p] where p is a point in some interval in X

Copyright © C. J. Date 2005page 81 EXPANDED FORM (cont.) : Given any such set X, a corresponding expanded form always exists; expanded form is equivalent to X and is unique Expanded form of X is one possible canonical form for X Unique set equivalent to X such that every contained interval is of minimum possible duration (viz., one) X1  X2 if and only if they have the same expanded form Intuitively, the expanded form of X allows us to focus on the information content of X at an atomic level, without worrying about the many different ways that information might be bundled together into clumps

Copyright © C. J. Date 2005page 82 COLLAPSED FORM : X1 = { [d01:d01], [d03:d05], [d04:d06] } X2 = { [d01:d01], [d03:d04], [d05:d05], [d05:d06] } X3 = { [d01:d01], [d03:d03], [d04:d04], [d05:d05], [d06:d06] } Expanded form here (X3) has greatest cardinality: fluke! X4 = { [d01:d01], [d03:d03], [d03:d04], [d03:d05], [d03:d06], [d04:d04], [d04:d05], [d04:d06] } X4 has same expanded form but greater cardinality than X3

Copyright © C. J. Date 2005page 83 COLLAPSED FORM (cont.) : X5 = { [d01:d01], [d03:d06] } X5 has same expanded form but minimum possible cardinality (it’s the collapsed form of X1, X2, X3, X4) If X is a set of intervals all of the same type, then the collapsed form of X is the set Y of intervals of the same type such that: l X and Y have the same expanded form, and l No two distinct intervals i1 and i2 in Y are such that i1 MERGES i2 is true

Copyright © C. J. Date 2005page 84 COLLAPSED FORM (cont.) : Given any such set X, a corresponding collapsed form always exists; collapsed form is equivalent to X and is unique Collapsed form of X is another possible canonical form for X Unique set equivalent to X that has the minimum possible cardinality X1  X2 if and only if they have the same collapsed form Intuitively, the collapsed form of X allows us to focus on the information content of X in a compressed (clumped) form, without worrying about the possibility that clumps might overlap or abut

Copyright © C. J. Date 2005page 85 LET X BE A SET OF INTERVALS ALL OF THE SAME TYPE : l EXPAND ( X ):returns expanded form of X l COLLAPSE ( X ):returns collapsed form of X (What happens if X has cardinality zero? Or one?)  Ops are not inverses of each other! In fact: l EXPAND ( COLLAPSE ( X ) )  EXPAND ( X ) l COLLAPSE ( EXPAND ( X ) )  COLLAPSE ( X )

Copyright © C. J. Date 2005page 86 NOW I NEED TO CLEAN UP MY ACT ! Relational model doesn’t support general sets, it supports relations! However, a set of values v1, v2, …, vn all of the same type can easily be converted into a unary relation: RELATION { TUPLE { A v1 }, TUPLE { A v2 }, …………………, TUPLE { A vn } } Returns:A relation v1selector v2invocation. vn

Copyright © C. J. Date 2005page 87 So let us replace EXPAND and COLLAPSE as previously described by versions in which argument is specified as a unary relation rEXPAND(r)COLLAPSE(r) DURINGDURINGDURING [d06:d09] [d01:d01] [d01:d01] [d04:d08] [d04:d04] [d04:d10] [d05:d10] [d05:d05] [d01:d01] [d06:d06] [d07:d07] [d08:d08] [d09:d09] [d10:d10] Extend definition of equivalence accordingly

Copyright © C. J. Date 2005page 88 EXPANDING / COLLAPSING NULLARY RELATIONS : Nullary relation has no attributes at all … There are exactly two such relations! l TABLE_DEE has just one tuple (the "0-tuple") l TABLE_DUM has no tuples at all Highly desirable to define versions of EXPAND and COLLAPSE for nullary relations (see later) Definition: Result = input in every case

Copyright © C. J. Date 2005page LAYING THE FOUNDATIONS :  Time and the DB  What’s the problem?  Intervals  Interval operators  The EXPAND and COLLAPSE operators  The PACK and UNPACK operators Relational operators

Copyright © C. J. Date 2005page 90 THE PACK AND UNPACK OPERATORS : l Preliminaries l The PACK operator l The UNPACK operator l Sample queries l Packing and unpacking on no attributes l Packing and unpacking on several attributes

Copyright © C. J. Date 2005page 91 PRELIMINARY EXAMPLE : UNPACK rPACK r rON DURINGON DURING S#DURINGS#DURINGS#DURING S2[d02:d04] S2[d02:d02] S2[d02:d05] S2[d03:d05] S2[d03:d03] S4[d02:d06] S4[d02:d05] S2[d04:d04] S4[d09:d10] S4[d04:d06] S2[d05:d05] S4[d09:d10] S4[d02:d02] S4[d03:d03] S4[d04:d04] S4[d05:d05] S4[d06:d06] S4 [d09:d09] S4[d10:d10]

Copyright © C. J. Date 2005page 92 RECALL QUERY A : l Get S#-FROM-TO triples for suppliers who have been able to supply at least one part during at least one interval of time, where FROM and TO together designate a maximal interval during which supplier S# was in fact able to supply at least one part Revised version: l Get S#-DURING pairs for suppliers who have been able to supply at least one part during at least one interval of time, where DURING designates a maximal interval during which supplier S# was in fact able to supply at least one part We will build up our formulation one step at a time...

Copyright © C. J. Date 2005page 93 S#P#DURING S1P1[d04:d10] S1P2[d05:d10] S1P3[d09:d10] S1P4[d05:d10] S1P5[d04:d10] S1P6[d06:d10] S2P1[d02:d04] S2P1[d08:d10] S2P2[d03:d03] S2P2[d09:d10] S3P2[d08:d10] S4P2[d06:d09] S4P4[d04:d08] S4P5[d05:d10] SP_DURING (SAMPLE VALUE) :

Copyright © C. J. Date 2005page 94 PROJECT AWAY PART NUMBERS : S#DURING S1[d04:d10]WITH SP_DURING { S#, DURING } S1[d05:d10]AS T1 : S1[d09:d10]/* part numbers irrelevant */ S1[d06:d10] S2[d02:d04] S2[d08:d10]Note the redundancy— S2[d03:d03]E.g., "Supplier S1 was able to S2[d09:d10]supply something on day 6" S3[d08:d10]appears three times! S4[d06:d09] S4[d04:d08] S4[d05:d10] T1

Copyright © C. J. Date 2005page 95 DESIRED RESULT (ELIMINATING REDUNDANCY) : S#DURING S1[d04:d10]Packed form S2[d02:d04]of T1 on DURING: S2[d08:d10] S3[d08:d10]"PACK T1 ON DURING" S4[d04:d10] Note:Given DURING value for given supplier in RESULT does not necessarily exist as an explicit DURING value for that supplier in T1 (see, e.g., S4)

Copyright © C. J. Date 2005page 96 WITH ( T1 GROUP { DURING } AS X ) AS T2 : S#XAttribute X is relation-valued! S1DURING [d04:d10] [d05:d10] [d09:d10]S3DURING [d06:d10] [d08:d10] S2DURING S4DURING [d02:d04] [d08:d10][d06:d09] [d03:d03][d04:d08] [d09:d10] [d05:d10] T2 Next..