postgresql/contrib/cube/README.cube
Tom Lane 31edbadf4a Downgrade implicit casts to text to be assignment-only, except for the ones
from the other string-category types; this eliminates a lot of surprising
interpretations that the parser could formerly make when there was no directly
applicable operator.

Create a general mechanism that supports casts to and from the standard string
types (text,varchar,bpchar) for *every* datatype, by invoking the datatype's
I/O functions.  These new casts are assignment-only in the to-string direction,
explicit-only in the other, and therefore should create no surprising behavior.
Remove a bunch of thereby-obsoleted datatype-specific casting functions.

The "general mechanism" is a new expression node type CoerceViaIO that can
actually convert between *any* two datatypes if their external text
representations are compatible.  This is more general than needed for the
immediate feature, but might be useful in plpgsql or other places in future.

This commit does nothing about the issue that applying the concatenation
operator || to non-text types will now fail, often with strange error messages
due to misinterpreting the operator as array concatenation.  Since it often
(not always) worked before, we should either make it succeed or at least give
a more user-friendly error; but details are still under debate.

Peter Eisentraut and Tom Lane
2007-06-05 21:31:09 +00:00

356 lines
11 KiB
Plaintext

This directory contains the code for the user-defined type,
CUBE, representing multidimensional cubes.
FILES
-----
Makefile building instructions for the shared library
README.cube the file you are now reading
cube.c the implementation of this data type in c
cube.sql.in SQL code needed to register this type with postgres
(transformed to cube.sql by make)
cubedata.h the data structure used to store the cubes
cubeparse.y the grammar file for the parser (used by cube_in() in cube.c)
cubescan.l scanner rules (used by cube_yyparse() in cubeparse.y)
INSTALLATION
============
To install the type, run
make
make install
The user running "make install" may need root access; depending on how you
configured the PostgreSQL installation paths.
This only installs the type implementation and documentation. To make the
type available in any particular database, as a postgres superuser do:
psql -d databasename < cube.sql
If you install the type in the template1 database, all subsequently created
databases will inherit it.
To test the new type, after "make install" do
make installcheck
If it fails, examine the file regression.diffs to find out the reason (the
test code is a direct adaptation of the regression tests from the main
source tree).
By default the external functions are made executable by anyone.
SYNTAX
======
The following are valid external representations for the CUBE type:
'x' A floating point value representing
a one-dimensional point or one-dimensional
zero length cubement
'(x)' Same as above
'x1,x2,x3,...,xn' A point in n-dimensional space,
represented internally as a zero volume box
'(x1,x2,x3,...,xn)' Same as above
'(x),(y)' 1-D cubement starting at x and ending at y
or vice versa; the order does not matter
'(x1,...,xn),(y1,...,yn)' n-dimensional box represented by
a pair of its opposite corners, no matter which.
Functions take care of swapping to achieve
"lower left -- upper right" representation
before computing any values
Grammar
-------
rule 1 box -> O_BRACKET paren_list COMMA paren_list C_BRACKET
rule 2 box -> paren_list COMMA paren_list
rule 3 box -> paren_list
rule 4 box -> list
rule 5 paren_list -> O_PAREN list C_PAREN
rule 6 list -> FLOAT
rule 7 list -> list COMMA FLOAT
Tokens
------
n [0-9]+
integer [+-]?{n}
real [+-]?({n}\.{n}?|\.{n})
FLOAT ({integer}|{real})([eE]{integer})?
O_BRACKET \[
C_BRACKET \]
O_PAREN \(
C_PAREN \)
COMMA \,
Examples of valid CUBE representations:
--------------------------------------
'x' A floating point value representing
a one-dimensional point (or, zero-length
one-dimensional interval)
'(x)' Same as above
'x1,x2,x3,...,xn' A point in n-dimensional space,
represented internally as a zero volume cube
'(x1,x2,x3,...,xn)' Same as above
'(x),(y)' A 1-D interval starting at x and ending at y
or vice versa; the order does not matter
'[(x),(y)]' Same as above
'(x1,...,xn),(y1,...,yn)' An n-dimensional box represented by
a pair of its diagonally opposite corners,
regardless of order. Swapping is provided
by all comarison routines to ensure the
"lower left -- upper right" representation
before actaul comparison takes place.
'[(x1,...,xn),(y1,...,yn)]' Same as above
White space is ignored, so '[(x),(y)]' can be: '[ ( x ), ( y ) ]'
DEFAULTS
========
I believe this union:
select cube_union('(0,5,2),(2,3,1)','0');
cube_union
-------------------
(0, 0, 0),(2, 5, 2)
(1 row)
does not contradict to the common sense, neither does the intersection
select cube_inter('(0,-1),(1,1)','(-2),(2)');
cube_inter
-------------
(0, 0),(1, 0)
(1 row)
In all binary operations on differently sized boxes, I assume the smaller
one to be a cartesian projection, i. e., having zeroes in place of coordinates
omitted in the string representation. The above examples are equivalent to:
cube_union('(0,5,2),(2,3,1)','(0,0,0),(0,0,0)');
cube_inter('(0,-1),(1,1)','(-2,0),(2,0)');
The following containment predicate uses the point syntax,
while in fact the second argument is internally represented by a box.
This syntax makes it unnecessary to define the special Point type
and functions for (box,point) predicates.
select cube_contains('(0,0),(1,1)', '0.5,0.5');
cube_contains
--------------
t
(1 row)
PRECISION
=========
Values are stored internally as 64-bit floating point numbers. This means that
numbers with more than about 16 significant digits will be truncated.
USAGE
=====
The access method for CUBE is a GiST index (gist_cube_ops), which is a
generalization of R-tree. GiSTs allow the postgres implementation of
R-tree, originally encoded to support 2-D geometric types such as
boxes and polygons, to be used with any data type whose data domain
can be partitioned using the concepts of containment, intersection and
equality. In other words, everything that can intersect or contain
its own kind can be indexed with a GiST. That includes, among other
things, all geometric data types, regardless of their dimensionality
(see also contrib/seg).
The operators supported by the GiST access method include:
a = b Same as
The cubements a and b are identical.
a && b Overlaps
The cubements a and b overlap.
a @> b Contains
The cubement a contains the cubement b.
a <@ b Contained in
The cubement a is contained in b.
(Before PostgreSQL 8.2, the containment operators @> and <@ were
respectively called @ and ~. These names are still available, but are
deprecated and will eventually be retired. Notice that the old names
are reversed from the convention formerly followed by the core geometric
datatypes!)
Although the mnemonics of the following operators is questionable, I
preserved them to maintain visual consistency with other geometric
data types defined in Postgres.
Other operators:
[a, b] < [c, d] Less than
[a, b] > [c, d] Greater than
These operators do not make a lot of sense for any practical
purpose but sorting. These operators first compare (a) to (c),
and if these are equal, compare (b) to (d). That accounts for
reasonably good sorting in most cases, which is useful if
you want to use ORDER BY with this type
The following functions are available:
cube_distance(cube, cube) returns double
cube_distance returns the distance between two cubes. If both cubes are
points, this is the normal distance function.
cube(float8) returns cube
This makes a one dimensional cube with both coordinates the same.
If the type of the argument is a numeric type other than float8 an
explicit cast to float8 may be needed.
cube(1) == '(1)'
cube(float8, float8) returns cube
This makes a one dimensional cube.
cube(1,2) == '(1),(2)'
cube(float8[]) returns cube
This makes a zero-volume cube using the coordinates defined by the
array.
cube(ARRAY[1,2]) == '(1,2)'
cube(float8[], float8[]) returns cube
This makes a cube, with upper right and lower left coordinates as
defined by the 2 float arrays. Arrays must be of the same length.
cube('{1,2}'::float[], '{3,4}'::float[]) == '(1,2),(3,4)'
cube(cube, float8) returns cube
This builds a new cube by adding a dimension on to an existing cube with
the same values for both parts of the new coordinate. This is useful for
building cubes piece by piece from calculated values.
cube('(1)',2) == '(1,2),(1,2)'
cube(cube, float8, float8) returns cube
This builds a new cube by adding a dimension on to an existing cube.
This is useful for building cubes piece by piece from calculated values.
cube('(1,2)',3,4) == '(1,3),(2,4)'
cube_dim(cube) returns int
cube_dim returns the number of dimensions stored in the the data structure
for a cube. This is useful for constraints on the dimensions of a cube.
cube_ll_coord(cube, int) returns double
cube_ll_coord returns the nth coordinate value for the lower left corner
of a cube. This is useful for doing coordinate transformations.
cube_ur_coord(cube, int) returns double
cube_ur_coord returns the nth coordinate value for the upper right corner
of a cube. This is useful for doing coordinate transformations.
cube_subset(cube, int[]) returns cube
Builds a new cube from an existing cube, using a list of dimension indexes
from an array. Can be used to find both the ll and ur coordinate of single
dimenion, e.g.: cube_subset(cube('(1,3,5),(6,7,8)'), ARRAY[2]) = '(3),(7)'
Or can be used to drop dimensions, or reorder them as desired, e.g.:
cube_subset(cube('(1,3,5),(6,7,8)'), ARRAY[3,2,1,1]) = '(5, 3, 1, 1),(8, 7, 6, 6)'
cube_is_point(cube) returns bool
cube_is_point returns true if a cube is also a point. This is true when the
two defining corners are the same.
cube_enlarge(cube, double, int) returns cube
cube_enlarge increases the size of a cube by a specified radius in at least
n dimensions. If the radius is negative the box is shrunk instead. This
is useful for creating bounding boxes around a point for searching for
nearby points. All defined dimensions are changed by the radius. If n
is greater than the number of defined dimensions and the cube is being
increased (r >= 0) then 0 is used as the base for the extra coordinates.
LL coordinates are decreased by r and UR coordinates are increased by r. If
a LL coordinate is increased to larger than the corresponding UR coordinate
(this can only happen when r < 0) than both coordinates are set to their
average. To make it harder for people to break things there is an effective
maximum on the dimension of cubes of 100. This is set in cubedata.h if
you need something bigger.
There are a few other potentially useful functions defined in cube.c
that vanished from the schema because I stopped using them. Some of
these were meant to support type casting. Let me know if I was wrong:
I will then add them back to the schema. I would also appreciate
other ideas that would enhance the type and make it more useful.
For examples of usage, see sql/cube.sql
CREDITS
=======
This code is essentially based on the example written for
Illustra, http://garcia.me.berkeley.edu/~adong/rtree
My thanks are primarily to Prof. Joe Hellerstein
(http://db.cs.berkeley.edu/~jmh/) for elucidating the gist of the GiST
(http://gist.cs.berkeley.edu/), and to his former student, Andy Dong
(http://best.me.berkeley.edu/~adong/), for his exemplar.
I am also grateful to all postgres developers, present and past, for enabling
myself to create my own world and live undisturbed in it. And I would like to
acknowledge my gratitude to Argonne Lab and to the U.S. Department of Energy
for the years of faithful support of my database research.
------------------------------------------------------------------------
Gene Selkov, Jr.
Computational Scientist
Mathematics and Computer Science Division
Argonne National Laboratory
9700 S Cass Ave.
Building 221
Argonne, IL 60439-4844
selkovjr@mcs.anl.gov
------------------------------------------------------------------------
Minor updates to this package were made by Bruno Wolff III <bruno@wolff.to>
in August/September of 2002.
These include changing the precision from single precision to double
precision and adding some new functions.
------------------------------------------------------------------------
Additional updates were made by Joshua Reich <josh@root.net> in July 2006.
These include cube(float8[], float8[]) and cleaning up the code to use
the V1 call protocol instead of the deprecated V0 form.