| BindRules(7) | ShapeTools | BindRules(7) |
BindRules - ShapeTools version bind rules
The ShapeTools version binding subsystem (see vbind(1)) provides a mechanism for expressing general version bind rules. These rules describe on an abstract level version properties, which will be matched against the properties of concrete versions during the version bind procedure. The goal is to select one or more versions from a named history in order to provides access to these version(s). A version bind operation is always performed for exactly one history at a time. Version bind rules express something like
Select the most recent saved version. If there is no saved version, select the busy version.
ShapeTools however needs rules in a more formal notation to be able to interpret them. Let's see, how the rule above is translated into the formal notation.
Version bind rules consist of a list of attribute expressions evaluated one after another until one of the expressions leads to a unique version identification. The expressions are separated by semicolons, the last expression ends with a period. The rule from above will now read:
Select the most recent saved version ; Select the busy version .
Each attribute expression consist of a list of predicates, separated by commas. The predicates are evaluated from left to right resulting in a hit set, a set of versions fulfilling all predicates evaluated so far. The initial hit set for an attribute expression contains all versions of the name to be bound. Each predicate potentially narrows the hit set. The predicates in our rule are:
all saved versions , most recent version ; busy version .
Remember, that each predicate bases it's selection on the hit set left by the predicate before. Hence exchanging the two predicates in the first attribute expression may lead to different results. We will give more information on this topic in the section about the evaluation algorithm below. We now reach the final form of ShapeTools version bind rules. The predicates must be taken from a list of predefined names and be equipped with arguments:
ge (status, saved) , max (stime) ; eq (status, busy) .
That's it so far. This is a rule how ShapeTools understands it. It does however illustrate just a small piece of the world of version bind rules. We will go on in this manual page with a detailed description of version bind rules divides into the sections
A version bind rule consists of a rule head and a rule body. The example above shows only the rule body. The rule head defines a name for the rule and optionally a parameter list. The name is a string consisting of any printable non-whitespace character except colon and parentheses. It is followed by an optional parameter list in parentheses and a colon, delimiting the rule head. Multiple parameters in the list are separated by comma. Examples are
most_recently_released: from_release (release_name): last_released (library_path, include_path):
The basic idea of the rule evaluation algorithm is, that in every state of processing a hit set exists, a set of versions reflecting the current rule evaluation result. The hit set is initialized with all versions of the given name at the beginning of each attribute expression. The attribute expressions predicates are processed from left to right in the order they occur. Each predicate imposes requirements to the versions in the hit set and eliminates all versions not fulfilling these requirements. So, the hit set becomes smaller and smaller during attribute expression evaluation. The following figure illustrates this process together with the rule most_recently_released defined above and the file foo existing as busy version and as versions 1.0 through 1.2.
Initial hit set: ( foo[busy], foo[1.0], foo[1.1], foo[1.2] ) Evaluate Predicate: ge (status, saved), New hit set: ( foo[1.0], foo[1.1], foo[1.2] ) Evaluate Predicate: max (stime); Final hit set: ( foo[1.2] )
When the hit set becomes empty, that is when no version meets all the predicates evaluated so far, the attribute expression fails and processing is finished immediately. All remaining predicates will not be evaluated. Even remaining predicates without influence on the hit set (for example message predicates) will not be processed. Processing continues with the next attribute expression. If all attribute expressions finish prematurely, the whole version binding fails. In the following example, the first attribute expression fails and the second alternative leads to success.
Initial hit set: ( bar[busy] ) Evaluate Predicate: ge (status, saved), New hit set (empty): ( ) Evaluate next attribute expression starting with initial hit set again: ( bar[busy] ) Evaluate Predicate: eq (status, busy); Final hit set: ( bar[busy] )
When evaluation reaches the end of an attribute expression without the hit set being empty, two cases are possible. First, the hit set contains exactly one version and everything is fine. This is usually the desired state and rule evaluation returns the remaining version as bind result. Second, the hit set may contain more than one version. In this case, the evaluation algorithm depends on the expected result. When a unique version binding is expected, this is treated as failure and evaluation goes on with the next attribute expression. Non-unique version binding regards this as success and returns the whole hit set.
Extending the hit set during evaluation of an attribute expression is not possible. This would be against the nature of the version bind rules and would make them much more difficult to understand. Hit set extension may only happen by letting the current attribute expression fail and begin with a new one and the maximum hit set.
Failure of an attribute expression must not necessarily be caused by an empty hit set. It may also be caused by user interaction or by external constraints. The following rules exemplify user interaction:
eq (state, busy), confirm (select busy version ?, y); ge (state, busy), max (version).
where the user will be asked for confirmation to select the busy version, and external constraints:
exists (otto, 1.0), eq (state, busy); ge (state, busy), max (version).
where selection of the busy version happens only, when version 1.0 of otto exists (this example is somewhat silly). Predicates like confirm and exists don't care about the hit set. They provide the possibility to impose external control on the evaluation of version bind rules. An attribute expression may be finished prematurely and control switches to the next one.
There is another operator, the cut operator, that forces the whole bind operation to finish (and fail). Typically the cut operator stands at the end of an attribute expression that should never succeed. The following is a typical example for use of this. Version binding fails, if there is an update lock set on the most recent version.
max (version), hasattr (locker), cut (history is locked !); max (version).
The cut operator accepts a string argument that will be written to the standard output.
Each attribute expression may start with a pattern, against which the name to be bound is matched. Only when the name matches the pattern, the corresponding attribute expression will be evaluated. If not, the attribute expression will be skipped. When the pattern is omitted in the attribute expression (as in our example above), the expression is evaluated for each name.
The patterns are the same as those recognized by sh(1) for filename generation on the command line. Magic cookies are:
A rule with name patterns for example looks like:
xyyz.h, eq (version, 1.3); *.c, eq (generation, 2), max (revision); *.h, eq (generation, 3), max (revision).
In this example, version binding for C source files (most recent version from generation 2) is different from version binding for header files (most recent version from generation 3). Additionally, the name xyyz.h will always be bound to version 1.3.
If the name to be bound is given together with a (absolute or relative) pathname, this will not be cut off. The match is always performed lexically for the whole name given. Hence, the name pattern may also contain path names, like
variant1/*, eq (alias, var1-1.4); variant2/*, eq (alias, var2-1.2); /usr/sample/include/*.h, max (revision).
Usually, the version bind subsystem does not check, if different path prefixes in the name pattern and the given name to be bound lead to the same location. The match is done lexically and must fit exactly. An exception is, when the name pattern is given as network path name as in atnetwork(3). A network pathname consists of the name of the host, controlling the device where a version is stored, the canonical pathname to the version and a version binding (e.g. version number, version alias, or date) either in brackets or separated from the name by an at (@) sign. Examples are
desaster:/usr/sample/project/foo.c[1.0]; desaster:/usr/sample/project/variant1/bar.c[var1-1.4]; desaster:/usr/sample/project/xyyz.h@1.3; desaster:/usr/sample/project/bar.c@Fri Jun 18 13:40:58 MET DST 1993.
Network pathnames are mapped to canonical local pathnames before being processes and in this case, the given name to be bound will also be mapped to a canonical local pathname.
The technique using network pathnames is especially useful when storing the result of a successful version selection persistently. This makes the version selection easily reproducible from anywhere in the local areas network. shape(1) uses this mechanism when generating its bound configuration threads.
This is the complete list of valid predicate names and a synopsis of their arguments. The list is divided into several parts, each describing a certain class of predicates.
The first class are predicates working independently on each element of the current hit set. They impose certain requirements to the attributes of each version and eliminate those, not fulfilling the requirements.
The second class are predicates that do not operate on single version but rather on the complete hit set. They express relations between different versions in the hit set and base their selection on comparison of different versions. Usually, they are used to get a unique version binding, by ordering the hit set and selecting one of the extremes.
The next two predicate groups have no direct influence on the hit set. They can invalidate the hit set and cause the rule evaluation to go on with the next attribute expression, but they do never modify the hit set. These predicates are activated, when the evaluation of the attribute expression reaches them, i.e. when the hit set is not empty.
The last predicate group are external constraints. Their task is to influence the evaluation process by examining conditions outside the handled version history. Each of the following predicates has either a positive or a negative result. Positive results have no effect on the hit set and the evaluation process, while negative results invalidate the hit set and cause evaluation to go on with the next attribute expression.
There are a number of known predicate names from former versions of the bind rules machinery. They are internally mapped to the new predicate names. These names are obsolete and should not be used any longer.
Obsolete name mapped to - cut attr eq attrex hasattr attrge ge attrgt gt attrle le attrlt lt attrmax max attrmin min attrnot ne condex exists condnot existsnot conduniq existsuniq
All predicates with effect on the contents of the hit set work on version attributes. These attributes are either standard attributes with a defined meaning or user defined attributes. The following is a list of attribute names recognized as standard attributes. All other names are considered to be user defined attributes.
Some predicates (like ge or max) require an ordering relationship between attribute values. For user defined attributes, ordering bases on alphabetical (ASCII) comparison of the values. User defined attributes with multiple values are compared by their first values, if these are identical by their second values and so on. No value is considered smaller than any other value. For example
attr1 = anton attr2 = berta berta is smaller than anton karl
but
attr1 = anton attr2 = anton berta is bigger than berta karl
For some of the standard attributes listed above, we need special regulations.
are ordered by generation number first and revision number secondary (e.g. 1.2 is smaller than 2.1). Busy is smaller than any version number.
are ordered by the version numbers (see above) of the identified versions.
are ordered by simple text comparison. This has the effect that the youngest cache key is considered the biggest.
are ordered in the sequence as listed above. Busy is the lowest and frozen the highest state.
The order of user attributes bases on alphabetical comparison of the string username@domain.
Time comparison assumes older dates to be smaller than newer ones.
During evaluation of version bind rules, four different kinds of expansion are possible. These are parameter substitution, attribute expansion, external macro expansion and command substitution. Expansion happens, when a magic pattern is found in the rule text, starting with either a dollar sign ($) or, in case of command substitution, with a backward quote character (`).
Generally, expansion in version bind rules happens only within patterns and within predicate arguments. Bind rule syntax or predicate names cannot be introduced by substituted strings. Expansions outside patterns and predicate arguments are ignored and usually lead to an error message.
A parameter substitution string is usually introduced by the pattern $_ followed by the parameter name (an exception is $+ as shown below). The parameter name is optionally delimited by a dollar sign. This is necessary, when there is no whitespace character following. The parameter name may be any of the names specified in the rule head or one of the following predefined names.
A parameter may have the same name as a citeable attribute (see below). In this case, the parameter citation hides the attribute citation. There is no way to cite the value of an attribute when there is an equally named rule parameter. The reserved names rule, target, and hits are not allowed as parameter names.
An attribute expansion string looks exactly like a parameter substitution string. It is introduced by the pattern $_ followed by the attribute name which is optionally delimited by a dollar sign, when a non-whitespace character follows immediately. Attribute names may be built of any printable characters except '#'. Besides, it makes no sense to cite attributes with an equal sign ('=') in the attribute name, as the Attributed Filesystem (AtFS) doesn't allow this.
The value by which the attribute expansion string will be replaced depends on the current state of processing. This may cause different values to be inserted for the same citation in different processing states. Attribute expansion happens as late as possible, it is done right before the evaluation of the concerned pattern or predicate. With one exception, $_hits$, attribute expansions will only be substituted, if the current hit set cardinality is 1.
Attribute citations may be overloaded by parameter citations (see above).
External macros are evaluated by an external macro processor. If no such macro processor is available, external macros remain unchanged. They have the form:
A command enclosed in back quotes occuring in a bind rule quotes will be replaced by its output. No modifications are done to the command output, hence it may contain newline characters.
There are some characters with special meaning when occurring in version bind rules. These are the syntactical characters colon (:), comma, (,), semicolon (;), period (.), and parentheses (( and )), the comment symbol (#), the dollar sign ($) or the back quote (`) introducing expansion strings ($), quotes (" and '), and the escape symbol (\).
Comments are handled somewhat rigorously. A comment symbol (#) occurring anywhere in the rule name or rule body has effect as long as it is not escaped by a backslash (\) character. Comments range from the comment symbol (inclusive) to the end of the line. Newline characters within comments may also be escaped by a backslash character, continuing the comment on the next line.
Nesting of parentheses and quotes is not supported.
The following is a list of lexical constraints for each part of a version bind rule.
Why doesn't the bind rule select version xyz although I think it should ?. An important facility to find an answer to this question is the trace option provided by the vbind(1) command. It shows the evolution of the hit set during rule evaluation.
Typing errors in standard attribute names may lead to confusing situations. They cannot be recognized by the evaluation machinery, as any unknown attribute name is considered to be an user defined attribute.
A minus sign (-) as first character in an alternative is
considered as part of the pattern and not as (old style) cut
operator. Hence
-; (- as pattern)
and
,-; (default pattern followed by cut)
make a big difference. We recommend the use of cut() in any case.
The short form (-) is supported only for compatibility with older
versions.
bind_rule ::= bind_rule_head :[-] bind_rule_body . bind_rule_head ::= rule_name | rule_name ( rule_arg_list ) . rule_arg_list: rule_name { , arg_name }* . bind_rule_body ::= attr_expression { ; attr_expression}* . . attr_expression ::= name_pattern { , predicate }* | predicate { , predicate }* | . name_pattern ::= { <any printable character or whitespace> }+ predicate ::= attr_value_predicate ( attr_name , string ) | attr_name_predicate ( attr_name ) | bind_rule_predicate ( rule_name ) | msg_predicate ( string ) | msg_answer_predicate ( string , string ) | cond_rule_predicate ( string , bind_rule_head ) | cond_expr_predicate ( string , string ) | cut_predicate. attr_value_predicate ::= eq | ge | gt | le | lt | ne . attr_name_predicate ::= hasattr | max | min . bind_rule_predicate ::= bindrule . msg_predicate ::= cut | msg . msg_answer_predicate ::= confirm . cond_rule_predicate ::= exists | existsnot | existsuniq . cond_expr_predicate ::= condexpr . cut_predicate ::= - . attr_name ::= arg_name | author | atime | ctime | generation | locker | ltime | mtime | owner | revision | size | state | stime | version . rule_name ::= { <any printable character except colon and parentheses }+ . arg_name ::= { <any printable character except comma, colon and parentheses }+ . string ::= { <any printable character or whitespace> } .
$SHAPETOOLS/BindRules
Andreas.Lampen@cs.tu-berlin.de
| Tue Jun 29 16:16:30 1993 | AtFStk-1.12 |