Keywords in Oracle
February 11, 2008
Some words in Oracle have special meaning. Maybe not to you, but then youre not the one having to parse, execute and fetch what you type. To be more precise, the some words in the first sentence can be categorized as reserved words and keywords. Within the keywords category, context comes into play because a word isnt always reserved. For example, the word COMMIT by itself triggers many events, so youd probably assume COMMIT is one of those words that Oracle has a death grip on. From Oracles perspective, no one but me can use that word. As it turns out, COMMIT isnt held that closely. You can create a table named COMMIT if you so desire because COMMIT is a keyword, a lower classification than that of a reserved word.
A reserved word is locked down, whereas a keyword can be used under certain conditions. Auditing is a valuable tool or feature, and if you wanted to create your own audit table, could you issue a create table audit (...) statement?
At least in SQL, no, you may not use audit that way. So, how can you tell (or where can you find out) what the special words are, since those are clearly the ones you dont want to step on? Several guides within the documentation library (typically in an index) contain lists, but the one definitive and one-stop source is the V$RESERVED_WORDS data dictionary view.
At a high level, the name of the view suggests that this is only about reserved words; but when describing the view, the key column of interest is named KEYWORD. That seems confusing given that I just went into detail about the difference between a reserved word and a keyword. It turns out that a second column in the view also matters: RESERVED. The decoder ring for V$RESERVED_WORDS then is this:
The Database Reference guide shows exactly that meaning within the description of V$RESERVED_WORDS. The view has several other columns, and what they reveal (or not) is interesting. Take LENGTH for example. The column, as the name suggests, represents the length of the keyword. This raises two questions. First, why does the length matter? Second, why is the length stored as an attribute when it can be easily derived? (Which normal form does that violate?)
In Oracle 10g, there are 27 keywords with length of one, and 29 with length one or null. What are the one or less length keywords (and their reserved status)? Im considering length one because that will generally include symbols. The keyword of >> could be included as well, but then the number of words to consider increases beyond whats of interest here.
There are two instances of a null or blank value in the non-reserved keyword column. Do they represent a space, null or something else? Hard to tell since neither has an ASCII value. If one represented some form of whitespace, perhaps wed see an 8, 9, 10, or 32 for a backspace, tab, new line, or space, respectively.
Lets dig deeper into the view definition and see where the values come from. Using Toad and drilling down into the script for GV_$RESERVED_WORDS (under the SYS schema, VIEWS), we see the following:
SELECT inst_id, keyword, LENGTH, DECODE (MOD (TRUNC (TYPE / 2), 2), 0, 'N', 1, 'Y', '?') reserved, DECODE (MOD (TRUNC (TYPE / 4), 2), 0, 'N', 1, 'Y', '?') res_type, DECODE (MOD (TRUNC (TYPE / 8), 2), 0, 'N', 1, 'Y', '?') res_attr, DECODE (MOD (TRUNC (TYPE / 16), 2), 0, 'N', 1, 'Y', '?') res_semi, DECODE (MOD (TRUNC (TYPE / 32), 2), 0, 'N', 1, 'Y', '?') duplicate FROM x$kwddef;
After formatting some columns and ordering by type (only for length <= 1), the output is:
ADDR INDX INST_ID KEYWORD LENGTH TYPE -------- ---------- ---------- ------- ------- ------ 607DE010 1140 1 0 1 607DD48C 1073 1 E 1 1 607DC8B0 1004 1 A 1 1 607D6818 442 1 G 1 1 607D3B3C 181 1 M 1 1 607DC070 956 1 K 1 1 607DA230 780 1 U 1 1 607DE03C 1141 1 0 1 607D37A0 160 1 P 1 1 607DC40C 977 1 T 1 1 607D8E14 663 1 | 1 2 607D8FF8 674 1 : 1 2 607DA364 787 1 - 1 2 607DAA70 828 1 [ 1 2 607DADE0 848 1 < 1 2 607DBB74 927 1 ] 1 2 607DBE8C 945 1 ! 1 2 607DC82C 1001 1 * 1 2 607D8BAC 649 1 > 1 2 607D7D10 564 1 ( 1 2 607D7108 494 1 = 1 2 607D6CBC 469 1 . 1 2 607D6978 450 1 @ 1 2 607D4FDC 301 1 ^ 1 2 607D45B8 242 1 / 1 2 607D3A34 175 1 , 1 2 607D2B6C 89 1 ) 1 2 607D1F38 18 1 + 1 2 607D2300 40 1 & 1 2
The TYPE column is used as a bucket for grouping words. Overall, the counts of types are:
SQL> select type, count(*) 2 from x$kwddef 3 group by type 4 order by 1; TYPE COUNT(*) ---------- ---------- 1 998 2 95 9 2 16 29 33 14 34 4
Using the value of type as an indicator, designating a word as a duplicate means having a type value high enough (above 32) so that the truncation of the division is above one. The 18 duplicate-valued words are:
KEYWORD --------------------- PRIVILEGE NOPARALLEL_INDEX NESTED_TABLE_SET_REFS INTEGER NOREWRITE REFERENCING NO_FILTERING SB4 UB2 INDEX_RS SMALLINT NOCPU_COSTING PARALLEL ROLES MAXARCHLOGS CONSTRAINTS DECIMAL CHAR
Are these words really duplicates of other keywords? By count, the answer is no. There is only a difference of two between the count of distinct words (1140) and total words (1142). Is it somehow related to the purpose or function of the word, that is, is CHAR a (rough) duplicate of VARCHAR2? If that were the case, then VARCHAR (or VARCHAR2, pick one) should be a duplicate as well. The meaning of duplicate, then, is somewhat of a mystery.
Lets look at the problem from Oracles (the company) perspective. A technical writer on an OTN forum posting acknowledged a key issue behind creating new keywords. Many, if not all, documented optimizer hints are keywords of one form or another. Suppose you create a procedure named GO_FAST but then in the next release of the RDBMS, a new (and we waited such a long time for this!) hint named GO_FAST is now at your disposal. You can begin to imagine the howls of complaints from customers when established object names have to be renamed in order to not conflict with reserved words in a new release.
Nonetheless, the number of (at least) documented words has increased from 660 in 18.104.22.168 to 1142 in 10.2.0.3. What turns out when minusing the two views is that 10g has 487 keywords not found in 8i, and 8i has two (not counting the blanks) not found in 10g. Many of the newer words are related to optimizer hints.
As a developer, you have to be clever enough to pick a scheme that is unlikely to be used by users. Alternatively, a scheme could be established in the beginning of your RDBMS product that not only prevents users in general from conflicting with it, but also only allows a few select users to even see it in the first place. The X$ tables come to mind for this.
I try to decipher the hidden meaning behind X$ table names. Several sources have defined many of them; one fairly extensive source is Note 175982.1 on MetaLink. Even with that list, the meaning of kwddef is not shown. We know that the leading k is for kernel, and Id guess that wddef could be something related word definition.
The key is that what is named in SYS is very safe from conflict, but what is exposed (via views, as one way) in SYS risks conflicting with your code or naming scheme. Youve no doubt run across something like this already. At certain levels within Oracle, there can be one and only one name for an object. Users A and B can each have a table named EMP, but there can be at most one and only one public synonym with the same name that refers to EMP. Consider a user named A:
SQL> create table a.emp as select * from scott.emp; Table created. SQL> create public synonym emp for scott.emp; Synonym created. SQL> create public synonym emp for a.emp; create public synonym emp for a.emp * ERROR at line 1: ORA-00955: name is already used by an existing object
Establishing a naming convention, particularly with respect to keywords, reminds me of the observation made by the knight guarding the Holy Grail (in Indiana Jones and the Last Crusade) when Donovan (Jones nemesis) drank from the wrong goblet and dies a horrible death: He chose...poorly. Once keywords are chosen and a product is released to users, adding more keywords (or changing a non-reserved work into a reserved one) becomes problematic. Seemingly harmless choices or decisions early on may limit your choices in the future. Although allowing certain words to be redefined may seem like youre building in a lot of flexibility, you may just be adding to confusion down the road.
To further complicate matters, some of Oracles keywords are also ANSI reserved words. Moreover, if dealing with different systems, a keyword in system A is likely not going to be a keyword in system B. Later on in the movie, after Indiana picked the correct goblet, the knight says, You have chosen wisely. That may be hard to do when given numerous words to choose from. Pick what you need now, and if possible, set aside some reserved reserved words just in case. Its always easier to let them go than to try and claim them in the future.