Data are represented by values. Values are immutable; they are not changed by computation. A constant always denotes the same value within a scope. A variable is a value that may contain another value; assignment changes the value contained by a variable, but not the value that is the variable.

A value used in a program may be represented by a literal constant, the name of a constant or variable, or by an expression, which will itself contain other values. Every name occurring in the program must be introduced by a declaration. A declaration associates with a name both a data type and a constant value (which may itself be a variable, and contain different values at different times).

A data type defines both a set of values and the actions that may be performed on elements of that set. It may either be directly described in a declaration that uses it, or it may be referenced by a type name, introduced in a type declaration. The type of every constant, variable, and expression can be deduced from static analysis. This analysis is performed by the compiler to ensure that all programs are type-correct; thus the language is said to be strongly typed.

An enumerated type definition indicates an ordered set of values, i.e., introduces names standing for each value in the set. The simple types are the enumerated types, the subrange types, and the built-in types, including BOOL, INT, CARD, REAL, and CHAR. There are standard denotations for literal constants of the built-in types: TRUE and FALSE for BOOL, numbers for INT and its subranges and for REAL, quotations for CHAR. Numbers and quotations are syntactically distinct from names—as are the ``reserved words'' of the language. The set of values of type CHAR is an 8-bit variant of the ASCII character codes.

A type may be defined as a subrange of a simple type by indicating the smallest and largest value of the subrange.

Structured types are defined by describing the types of their components, and indicating a structuring method: ARRAY or RECORD. These differ in the mechanism for selecting a component of a value.

In an array structure, all components are of the same type. A component is selected by a computable selector, or index. The index type, which must be simple, is indicated in the array type definition. It is usually a programmer-defined enumerated type, or a subrange of INT. Given a value of the index type, an array selector yields a value of the component type. Every array structure value can therefore be regarded as a mapping of the index type into the component type.

In a record structure, the components (called fields) are not necessarily of the same type. In order that the type of a selected component be evident from the program text (without executing the program), a record selector is not a computable value, but must instead be a name uniquely denoting the component to be selected.

A record type may be specified as consisting of several variants. This allows different record values of the same type to have structures that differ in the number of components, their types, or their names. The variant describing a particular value is indicated by a special field, called its tag. Variants of a type may also share fields in addition to the tag.

An explicit variable declaration associates a name and a static variable; the name is used to denote the variable in expressions. Dynamic variables are generated by a special procedure (NEW) that yields a pointer or reference value that subsequently serves in place of a name to refer to the variable. Finite graphs in their full generality may be represented using pointers or references.

The simplest statement is the assignment statement. It specifies that a newly computed value be assigned to a variable (or a component of a variable). The value is obtained by evaluating an expression. Expressions consist of variables, constants, operators, and procedure values operating on arguments to produce new values. Constants are literal or declared; variables and procedures are built-in or declared; the set of operators is defined within the language, and includes operators for arithmetic, comparison, and logical operations.

The procedure statement causes the application (invocation, call) of a designated procedure value to the values of its arguments (actual parameters).

Basic statements are the components of structured statements, which specify sequential, selective, or repeated execution of their components. Sequential execution of a sequence of statements is specified by separating them by semicolons; conditional or selective execution by the if statement and the select statement; and repeated execution by loop statements.

A block can be used to associate declarations with statements. The names so declared have significance only within the block. Hence, the block is the scope of these names, and they are said to be local to the block. Since a block may appear as a statement, scopes may be nested.

A block can be the body of a procedure value. A procedure has a fixed number of parameters, each of which is denoted within the procedure by a name called the formal parameter. Actual argument values are supplied for parameters at each application. All input parameters are value parameters.

Procedures may also have results; applications of such procedures may appear within expressions.

Mesa modules are a ``programming in the large'' mechanism for partitioning a system into manageable units. They can be used to encapsulate abstractions, to provide a degree of protection, and to enforce ``information hiding.'' They are also the units of separate compilation.

There are two kinds of modules: DEFINITIONS modules, which define interfaces, and PROGRAM modules, which contain the executable code to implement these interfaces.

Definitions (or defs) modules define interfaces to abstractions. They typically declare some shared types, useful constants, and the domains and ranges of a set of procedure names. They compile into symbol tables, which are shared by both clients and implementations. Checks are performed when modules are bound into a configuration to ensure that separately compiled pieces have used consistent versions of the shared definitions. Interfaces produce no executable code; they manifest themselves at runtime primarily as symbol tables that are accessible for debugging and similar purposes.

Program modules provide implementations of abstractions. They typically declare collections of variables that define their state and provide bodies for the procedures of their interfaces. Viewed as source text, they are similar to Pascal procedures and Simula class definitions. They can be loaded and interconnected to form complete systems.

This is still true, but I don't thinka beginner ought to know that. At runtime, one or more instances of an implementation may be created. A separate global frame (activation record) is allocated for each, containing storage for its global variables (those which are declared outside its procedures), which persist between applications of its procedures. The lifetimes of implementation instances typically (but not always) starts when the module is loaded and will remain (typically) as long as the program runs. The module body itself generally contains the code to initialize the global variables and establish any necessary invariants. It will be executed when the module is started, or upon application of one of the module's procuedures, whichever comes first.

A module that accesses (relies on declarations from) other modules must include DIRECTORY statements, so the necessary symbol tables can be acquired. If it uses only a subset of the declarations, it is good practice to indicate which ones with a USING list. Declarations in an interface are public unless declared to be PRIVATE. Normally the importing module accesses only the public names; private declarations may be accessed by implementing modules that indicate they SHARE the interface. A directory statement may list the name of a file containing the symbol table to be used, but if the file name is the same as the module name (except for the extension) it is omitted.

A module that uses non-constant declarations (e.g., exported types and procedures) from another module must explicitly import it. If a module implements any part of an interface (e.g., by supplying the value of a procedure or type that it declares), it must explicitly export it. The compiler will check that its PUBLIC declarations are type-consistent with the corresponding declarations in the exported interface(s).

Each module is effectively parameterized by a set of interface records, one for each interface it imports, and supplies a set of export records, one for each interface it exports. Note that interfaces and implementations need not be in one-to-one correspondence. Binding a group of modules together into a configuration involves assigning values from the export records to the corresponding fields in the interface records. There is a special sublanguage, C/Mesa, to control this process.

Accessing other modules introduces compilation order dependencies. Each module must be compiled after the modules it accesses (and recompiled if they change), since the compiler needs their symbol tables. But information does not flow in the other direction. Modules that are not accessed by others (virtually all implementations) may be freely recompiled without invalidating previous compilation and checking of any other modules.

Types, as well as procedures, can be declared opaquely in interfaces and subsequently bound to concrete values supplied by implementations. This makes the internal structure of the type invisible to clients of the interface, and ensures that there can be no compilation dependencies between the definition of the concrete type and the interface module. The definition of the type can be changed at any time without requiring recompilation of the interface or any clients of the interface.

Effective use of Mesa requires a thorough understanding of modules and their use. They have significantly influenced our program design and construction techniques.

Programs are almost never self-contained modules; the importation and re-use of existing code has all the advantages of theft over honest toilwithout the moral stigma. Considerable emphasis is laid on the careful design of interfaces, and on their documentation. Since it is only interface changes that force recompilation (or perhaps even rewriting) of client programs, it is important that interfaces remain stable for substantial periods, even while their implementations are undergoing change.

A recommended approach is to define, comment, and circulate for review, all of the interfaces in a (sub)system before writing any of the implementations. Interfaces play much the same role as ``program design languages'' in other environments, with the additional advantages of being precisely defined and mechanically enforced.

The Mesa language definition omits many of the features commonly expected in programming languages, such as input/output and string-manipulation operations. Of course, these facilities are available to Mesa programmers, but they are provided by packages written in the language itself. The descriptions of standard packages in the Mesa Programmer's Manual, Version 8.0, run to more than 300 pages.

When managing large collections of modules (and in systems like the Mesa Development Environment and Cedar they run into the thousands), module names become very important. The use of cryptic or acronymic names is discouraged. By convention, source file names have the extension .mesa, and object file names have a machine dependent extension (It used to be .bcd for Binary Configuration Description). The definitions module for an interface X is customarily named X; if it is implemented by a single program module, that implementation is customarily named XImpl.

Mesa provides a way to indicate when exceptional conditions arise in the course of execution and an orderly means for dealing with them that is inexpensive if they do not arise. Exceptions cause a transfer of control from the statement that raises them to a dynamically-selected part of the program intended to handle the situation. They may be raised in response to the detection of ``impossible'' situations, invalid inputs, the inability of an abstraction to supply its specified service, or simply unusual events.

Mesa exceptions are conceptually similar to procedures, except that the binding to the handler is determined by searching the catch phrases in the call stack of the process in which the exception is raised; the dynamically innermost handler that accepts the condition is applied. Like normal procedures, handlers can take parameters and return values. They are written in a distinctive syntax that clearly identifies them as code for the exceptional case.

Catch phrases are syntactically and semantically similar to SELECT statements, with test items indicating the exceptions for which the associated handler should be applied. There are special test items to catch arbitrary exceptions and to catch an attempt to unwind the application stack in response to an exception. A series of catch phrases may be associated with a procedure application, or enabled throughout a block.

A handler is like a procedure body, but when it completes, there are a number of additional control options: GOTO, EXIT, LOOP, RETRY, CONTINUE, REJECT, and RESUME. Resumption is analogous to returning from a procedure, possibly with a result. Exceptions are divided into SIGNALs, which may be resumed, and ERRORs, which may not; in common parlance they are generally all called signals.

Since handlers may take parameters and return results, each exception name must be declared in a scope that includes all the points where it is raised as well as all the catch phrases that accept it.

The cost of raising an exception is significantly higher than the cost of procedure application, but it shouldn't happen very often. The system guarantees that all exceptions are handled at some level; those that the program fails to catch are accepted by the debugger, keeping intact the state of the program that raised it.

Exceptions can be used in very intricate ways to achieve subtle effects (e.g., by raising another exception within a handler). Experience has shown that this is almost always a mistake. Some call it elegance, others call it incomprehensible:

Although his language proposals have not been implemented, The discussion in the working paper [Cedar]<Cedar7.0>Documentation>SignallingGuidelines.tioga is the best source of guidance on tasteful and appropriate uses of exceptions. The most important point is that the exceptions a procedure may raise must be considered part of its interface, and documented as such. Unfortunately, the compiler currently doesn't enforce this, and many otherwise excellent interfaces do not comply.

Mesa provides efficient mechanisms for concurrent execution of multiple processes within a single system. This makes it natural to structure programs to reflect their inherent concurrency. Mesa also provides facilities for mutually exclusive access to resources and process synchronization by means of entry to monitors and waiting on condition variables.

FORK makes it possible to start the execution of another procedure concurrently with the program that applies it. It returns a process, which may either be detached to proceed independently, or saved for a future JOIN. There is no rule against multiple coexisting instances of a procedure, either forked or applied, although care must be taken to ensure mutual exclusion on accesses to shared global data.

JOIN takes a single process argument. When the forked procedure has executed a RETURN and the JOIN has been executed (in either order), the returning process is deleted, and the joining process receives its results and continues execution. A process type is declared similarly to a procedure type, except that only the type of the result is specified.

All processes execute in the same address space. This means that they are not protected from each other, which is presumably acceptable in a single-user system. It also means that process creation and switching between processes is cheap (not much more time-consuming than a procedure call).

Generally, two or more cooperating processes need to interact in more complicated ways than simply forking and joining. The interprocess synchronization mechanism provided in Mesa is a variant of ``monitors'' adapted from the work of Hoare, Brinch Hansen, and Dijkstra. The underlying view is that interaction among processes is always based on access to shared resources (e.g., data) and that a proper vehicle for this interaction must unify the synchronization, the shared data, and the procedures that perform the accesses.

A monitor is typically a module instance, with shared data in its global frame, and its own procedures for accessing them. Some of the procedures are public, allowing applications of monitor procedures from outside. Obviously, conflicts could arise if two processes were executing in the same monitor at the same time. To prevent this, a monitor lock is used for mutual exclusion. Application of one of a monitor's ENTRY procedures automatically acquires its lock (waiting if necessary), and a return releases it. An integrity constraint that the programmer imposes on the monitor's data is called a monitor invariant. The lock makes it possible for the programmer to ensure that this invariant will be true whenever an entry procedure begins executionregardless of what is happening in various processessimply by making sure that it is true initially and that every entry procedure restores it before returning.

Of course, a process may enter the monitor and find that the monitor data is in a good state but indicates that the process may not proceed until some other process enters the monitor and changes the situation. The WAIT operation allows a process to release the monitor lock temporarily (and suspend execution) without returning. The wait is performed on a condition variable, which is associated by agreement with the actual condition needed. After making a change that may have changed the condition, some other process must perform a BROADCAST or NOTIFY on the condition variable; this allows a waiting process to reacquire the lock, retest the condition, and resume execution if it is true. Note that since a wait releases the lock, the monitor invariants must be restored before waiting.

The procedures of a monitor are classified as entry, internal, and external. Internal procedures may only be applied by entry or internal procedures of the same monitor, since they are intended to be executed within the monitor's mutual exclusion, but do not acquire the monitor lock. External procedures are logically outside the monitor, but are declared within the same module for reasons of logical packaging. Being outside, they must not reference any monitor data nor apply any internal procedures; they are often used to provide a convenient interface that ``hides'' one or more applications of entry procedures.

The attributes ENTRY and INTERNAL are associated with a procedure's body, not with its type; thus they do not appear in interfaces. From the client side of an interface, a monitor appears like any other module.

In simple cases, a monitor's data comprises its global variables, protected by an implicit lock that is automatically allocated in its global frame. However, many applications deal with multiple objects, represented, say, as records accessed through pointers. It may be necessary to ensure that operations on these objects are atomic, i.e., once the operation has begun, the object will not be otherwise referenced until the operation is finished. It is possible to associate a lock with the object, rather than with the module's global frame, by declaring the data as a MONITORED RECORD. A single module instance can then implement each operation as an entry procedure, taking the object as a parameter. Locking is specified in the module heading by a LOCKS clause.

A somewhat subtle source of deadlocks occurs if control leaves an entry procedure by means of an uncaught exception. Unless it is certain that all exceptions (including those raised by invoked procedures) are handled, each entry procedure should include an UNWIND catch phrase, which will implicitly release the monitor lock.

Mesa's facilities for ordinary sequential ``programming in the small'' are extensive, but fairly conventional. The syntax is not exactly like that of any other language, but for the most part it can be picked up easily with a few minutes study of the grammar. (In fact, since most program text is produced either by editing existing programs or by the use of the Tioga editor to expand syntactic templates, you may be able to just ``fake it.'') This section mentions a number of areas where Mesa provides ``convenience'' extensions or conceptually small changes.

SELECT statements generalize Pascal's ``case'' construct by allowing several ways to specify how one statement is to be chosen for execution from an ordered list. The most common form is based on the relation between the value of a given expression and those of expressions associated with each selectable statement. The relation may be equality (the default), any relational operator appropriate to the types of the values involved, or containment in a subrange. A single selection may be prefixed by several selectors, and an optional ENDCASE statement is selected only if none of the others are. Discriminating selection is used to branch on the type of a variant record value (and in Cedar, on the current type referred to by a REF ANY). SELECT expressions are analogous, but choose from an ordered list of expressions.

Iteration is provided by loop statements in which several different kinds of control can be freely intermixed. A loop has a control clause and a body. The control clause may specify a logical condition for normal termination, possibly combined with a range or a sequence of assignments for a controlled variable. In addition to ordinary statements, the body may contain EXIT or GOTO statements to explicitly terminate its execution, and may be followed by a REPEAT clause that acts like a selection on the GOTO used to terminate the loop. (GOTO cannot be used to synthesize arbitrary control structures. It is much more like a ``local'' exception.)

In Pascal, procedure execution must proceed somehow to the end of the body before terminating; in Mesa, it can be terminated anywhere by executing a RETURN statement. If the procedure's type includes results, the return statement may supply the values to be returnedotherwise they are taken from the result variables named in the type. Each procedure body is followed by an implicit return.

Pascal procedures are not values that may be assigned to variables; Mesa procedures are. In most cases, the programmer still thinks of a constant association between a procedure name and its body, but to truly understand what is going on when interface records are bound, it helps to realize that procedure values from the export records are being assigned to appropriate fields of the interface records. This same power is available to the Mesa programmer; one popular form of ``object-oriented programming'' is based on the creation of an explicit record of procedures for each kind of object, and passing around together a pair of pointers, one to the procedure record, and another to the object instance data.

INLINE procedure constants may be declared in interfaces or locally. This is an instruction to the compiler to expand the procedure body inline for each application, rather than compiling a call to out-of-line code. It is intended to improve the speed without changing the semantics of the procedureinlines are not macros. INLINE should be considered a form of tight binding best reserved for late stages of system tuning; among other things, it can cause the compiler to run out of resources, even when compiling what appear to be small modules.

Every expression in a Mesa program has a syntactic type that can be deduced from its structure by static analysis of the program text, a process called type determination. The language imposes constraints on the type of each expression according to the context in which it is used, even in separately compiled modules.

The type rules in Mesa take two general forms:

A sequence in Mesa is an indexable collection of items, all of which have the same type. In this respect, a sequence resembles an array; however, the length of the sequence is not part of its type. The (maximum) length of a sequence is specified when the object containing that sequence is created, and it cannot subsequently be changed. It is the responsibility of the programmer to keep track of the number of items in the sequence at any time.

Mesa allows a default initial value to be associated with a type. If a type is constructed from other types using one of Mesa's structures, such as RECORD, an implicit default value for the constructed type is derived from the default values of the component types, but it can be overridden with an explicit default value. Default values for arguments can simplify procedure applications; default fields of records make the corresponding constructors more concise and more convenient; initial values are useful to ensure that the corresponding storage is always well-formed, even before the variable has been used by the program.

Dynamic variables in Mesa are allocated in zones. These are not necessarily associated with fixed areas of storage; rather, they are objects characterized by procedures for allocation and deallocation. Mesa programs used to segregate allocations into different zones, however in Cedar nearly all allocations use the standard system zone. The operator NEW is used to create a dynamic variable in a zone, and FREE to release it; FREE is not used in Cedar.

The MACHINE DEPENDENT attribute allows precise control of the representation of values at the bit level.

Although Mesa pointers are typed, they provide a rich source of opportunities for creation of safety problems, including the classical dangling pointer problem, where a pointer is used after the storage it refers to has been deallocated, and the opposite storage leak problem, where storage becomes inaccessible without being deallocated for reuse. Freeing the programmer from responsibility for deallocating storage at just the right time was a major goal of Cedar. It adds a new class of REF types that are just like the corresponding pointer types except that the system is responsible for freeing the dynamic variables they refer to after they have become inaccessible.

The introduction of collectible storage has substantially revised programming style and interface design in Cedar. When the project was being contemplated, some Mesa programmers indicated that as much as 40% of their time went into designing and checking the code to avoid dangling pointers and storage leaks, to tracking errors in this code, and to wasting time in tracking other errors by suspecting storage deallocation problems. With REFs and a reliable garbage collector that all goes away.

A desirable property of a high-level language system is implementation independence. This means that the effects of (even erroneous) programs can be understood in terms of the languagerather than requiring an understanding of the particular implementation. Mesa comes rather close to meeting this goal (as evidenced by the fact that most Mesa debugging can be done ``at the Mesa level,'' without ever worrying about the format of frames or the details of storage management), but it does contain some unsafe features whose use can lead to messy implementation dependencies.

It was desirable on general grounds to reduce implementation-dependence in Cedar. However, the decision to include facilities for garbage collection made it imperative. A collector can cause storage to be deallocated (permitting its subsequent reallocation and re-use) at times that are completely unpredictable from examination of the source program. A single programming error that smashes a REF used by the collector can destroy data structures in ways that make it difficult to reconstruct any evidence of the original cause of the crash.

A major goal for the Cedar Language was that it contain a useful subset for which garbage collection would be safe. The safe subset of Cedar is basically that part of the language where even incorrect programs cannot interfere with the reliable operation of the collector. The vast majority of Cedar programs should be written primarily (or entirely) in the safe subset. Safe Cedar does not provide acceptably efficient substitutes for every use of Mesa's unsafe features, so Cedar provides a means for indicating that some regions of a program are trusted. This inhibits compiler enforcement of the safety restrictions and indicates that the programmer has assumed the additional responsibility of ensuring that these regions of the program do not violate the integrity of the system.

A desirable property of a high-level programming language is that is allow a wide range of binding times: that is, it should allow the programmer maximal control over when the attributes of a particular variable are determined, with different choices not requiring changes in all expressions containing the variable. Examples of such attributes are its type, storage allocation method, implementation (for abstract objects), and actual value; examples of binding times include program-writing time, compilation, configuration binding, program initialization, block entry, and statement execution. Generally speaking, deferring the binding of an attribute leads to greater generality in the program at the cost of decreased static checkability and (often) lower runtime efficiency.

Experience with languages like LISP and Smalltalk, in which most binding is done dynamically, shows that it is much easier to write certain kinds of programs, if type and/or implementation binding can be deferred. Programming tools (debuggers, performance monitors) and knowledge representation systems are typical examples. But few programs take full advantage of this flexibility very often. Cedar was designed to take advantage of early binding, as Mesa does, but to allow certain bindings to be explicitly deferred.

Although Cedar was not intended as a research project in programming languages, its developers were not immune to the temptation to make Mesa better in ways that were not strictly required to enable the new programming environment. This section discusses a few of these new features.

Let's suppose you want to run a simple program in Cedar. If an existing Mesa 5 or 6 program uses fairly vanilla stuff, it's easy to convert:

Here's what you need to do to your Mesa 5 or 6 program:

The changes in the Cedar language from Mesa 6 are fairly easy to understand for simple programs:

The @ operator is not permitted. There are three general ways to cope with this restriction: specializing, copying, and indirecting. For example, suppose you have a program that says

The form of variant record discrimination that does not copy the value to a new location cannot be used. Suppose you have a variant-record data structure like

Variant records cannot be overwritten. Similiar techniques can be used for sanitizing a program that overwrites variant records. Assuming the declarations of T and TR from above, suppose you wanted to write

Unsafe procedures cannot be passed as arguments to safe ones. The symptom of a violation of this rule is generally a message complaining about an incorrect type when there is no obvious type mismatch. All procedure types in an interface prefixed by CEDAR are implicitly prefixed with SAFE. The simplest thing to do is to put SAFE in front of PROC in the argument procedure declaration, and put TRUSTED in front of its body. As with all uses of TRUSTED, you should verify that the safety invariants are actually maintained, and document the reason for the TRUSTED in a comment.

This is a one-page reference grammar describing the complete syntax of the Cedar Language, in a compact variation on BNF developed by Butler Lampson. Keep it handy as you write programs. It provides a relatively compact source of information on the exact form of constructs accepted by the compiler. It will also alert you to much of the available variety in the languagebut of course, not every syntactically valid program makes semantic sense.

The parsing grammar used by the compiler is somewhat larger and more complex than the Reference Grammar. Some of this is for technical reasons associated with LALR(1) parsing, and some of it to enable the compiler to make certain semantic distinctions while parsing. The differences should be invisible when dealing with correct programs, but may affect the error messages given for incorrect ones.

This document contains four complete, runnable Cedar programs chosen to illustrate the use of most of the major features of the language, and to provide an introduction to the style of programming that is preferred in Cedar. You should certainly invest time in studying them before attempting to write Cedar programs. If you are one of those who learns best from examples, you may find them virtually the only tutorial information you need to learn the language.

These examples have been chosen so that they are also useful prototypes of kinds of programs you may want to write in Cedar. If you are like most Cedar programmers, you will probably find it easier to start from such a prototype, and change it to do what you want, than to enter a whole program ``from scratch.''

Because Cedar programmers so frequently read each other's code, it is considered good citizenship to adhere to certain stylistic conventions. Stylizing Cedar Programs discusses the generally agreed conventions.

You can save yourself a lot of typing, and produce nicely formatted code at the same time, by using Tioga's abbreviation expansion mechanism to generate all the high-level structure of your program (at least, all the bits that aren't simply copied). The file Cedar.abbreviations lists the available macros and their expansions; you can add your own favorites.

This is an annotated prototype that you will probably want to keep close to hand, because it compactly illustrates the most important principles from the previous document.

This is a precise definition of the complete syntax and semantics of the Cedar Language. It is available as a CSL blue & white report: Butler Lampson. A Description of the Cedar Language, PARC report CSL-83-15.

The formal definition of the language is given in terms of a kernel language, into which all Cedar constructs can be desugared to determine their precise semantics. The Reference Manual contains both the definition of the kernel, and an explanation of the desugarings. It also contains several tables that collect important information about the primitive types and type constructors of Cedar.

Be warned however that this is not only very hard to read but also doesn't describe the same language that the Cedar compiler understands. If you don't understand this manual, don't feel too bad, I don't understand it either.

This is intended to be the essence of the entire Cedar Language carefully condensed into two pages for ready reference. It covers both syntax and semantics, with examples and notes. It is definitely not for those with weak eyes, and should probably not even be read until you have studied the Reference Manual proper. But it should be very helpful in checking details that you may have forgotten. Keep it handy.

Since so much Cedar programming is done ``at the component level,'' you need to know what packages and tools are available and what they do. In general, full documentation (or at least the best available approximation thereto) for each component is stored on [Cedar7.0]<Documentation>, or is referenced in the component's DF file, stored on [Cedar7.0]<Top>.

The problem is finding out which components you should be interested in. That's where the Cedar Catalog, stored in CedarCatalog.tioga, comes in handy. It contains a somewhat structured list of all the components in Cedar considered ``interesting'' by their maintainers. A component may be interesting

For each entry, the Catalog indicates why it is considered interesting, and how to acquire documentation and the component itself. It also identifies the maintainer, who is the ultimate source of advice and help.

The Mesa Language Manual, Version 11.0, published by the Xerox Office Systems Division in 1984, is the most recent self-contained manual on the Mesa Language. It falls somewhere between a tutorial and a reference manual, and many users have complained that it isn't entirely satisfactory for either purpose. But if you need more information about the Mesa-like parts of Cedar, it may be your best source. However, this manual is very unsatisfactory as tutorial because it teaches the features but it is frequently very wrong on when to use and when to avoid any particular feature. Some of those mistakes are excusable as differences between Mesa and Cedar, others are caused by old fashioned ideas about memory usage of different features.

The newest version of the Mesa Language Manual I have seen carries version number 12. I can not say much about this manual as I haven't read it. However whatever the changes between Mesa 11.0 and Mesa 12.0 are, those changes have not made it into our compiler.

If you haven't managed to find information that you want after you have looked in what you consider to be the obvious places (or if you don't understand what you have found), don't hesitate to ask. Almost anyone in CSL is a fount of wisdom, willing to be asked almost any question on almost any subject. (Of course, the answers aren't equally reliable, but you can't have everything.) If the first person you ask doesn't know the answer, chances are good that you'll get a pointer to either a person or document that will have the answer.