Source Code 1992 March

home *** CD-ROM | disk | FTP | other *** search

/ Source Code 1992 March / Source_Code_CD-ROM_Walnut_Creek_March_1992.iso / unix_c / languags / icon / version7.doc < prev next >

Wrap

Text File | 1988-11-28 | 40.8 KB | 1,651 lines

Version 7 of Icon* Ralph E. Griswold Gregg M. Townsend Kenneth Walker TR 88-5a January 16, 1988 Department of Computer Science The University of Arizona Tucson, Arizona 85721 *This work was supported by the National Science Foundation under Grant DCR-8401831. Version 7 of Icon 1. Background The current version of Icon is Version 7. Version 7 contains a number of features not present in earlier versions, but other- wise is nearly the same as Version 5. The Icon book [1], which describes Version 5, continues to serve as the primary reference for Version 7. This report supplements the present Icon book. Together, the book and this report provide a complete description of Version 7. Most of the language extensions in Version 7 are upward- compatible with Version, 5 and almost all Version 5 and Version 6 Icon programs work properly under Version 7. New features are described in Section 2. However, some of the more implementation-dependent aspects described in the Version 5 book are now obsolete. See Section 3. 2. Features of Version 7 The major differences between Version 7 and Version 5 are listed below. Features that are new since Version 6 are identi- fied by squares ([]). Such designations are not precise, since there are many interactions between features in different ver- sions; they are intended only to focus attention. A declaration to allow the inclusion of separately translated Icon programs. A set data type. New options for sorting tables. Syntax to support the use of co-expressions in programmer-defined control operations. The invocation of functions and operators by their string names. [] New functions, especially for interfacing the operating system and manipulating bits. [] Correction of the handling of scanning environments. - 1 - [] Correction of the handling of co-expression return points and new co-expression facilities. [] Declaration of procedures with a variable number of argu- ments. [] The ability to convert potential run-time errors to expression failure. [] New keywords. [] Error traceback, which provides detailed information about the source of run-time errors. 2.1 The Link Declaration The link declaration simplifies the inclusion of separately translated libraries of Icon procedures. If icont [2] is run with the -c option, source files are translated into intermediate ucode files (with names ending in .u1 and .u2). For example, icont -c libe.icn produces the ucode files libe.u1 and libe.u2. The ucode files can be incorporated in another program with the new link declaration, which has the form link libe The argument of link is, in general, a list of identifiers or string literals that specify the names of files to be linked (without the .u1 or .u2). Thus, when running under UNIX*, link libe, "/usr/icon/ilib/collate" specifies the linking of libe in the current directory and col- late in /usr/icon/ilib. The syntax for paths may be different for other operating systems. The environment variable IPATH controls the location of files specified in link declarations. The value of IPATH should be a blank-separated string of the form p1 p2 ... pn where each pi names a directory. Each directory is searched in turn to locate files named in link declarations. The default value of IPATH is the current directory. The current directory is always searched first, regardless of the value of IPATH. 2.2 New Functions Most of the new functions are described in this section. See subsequent sections for functions specifically related to other *UNIX is a trademark of AT&T Bell Laboratories. - 2 - new features. [] System-Interface Functions getenv(s) getenv(s) produces the value of the environment variable s. It fails if the environment variable is not set. It also fails on systems that do not support environment variables. remove(s) remove(s) removes the file named s. Subsequent attempts to open the file fail, unless it is created anew. If the file is open, the behavior of remove(s) is system dependent. remove(s) fails if it is unsuccessful. rename(s1,s2) rename(s1,s2) causes the file named s1 to be henceforth known by the name s2. The file named s1 is effectively removed. If a file named s2 exists prior to the renaming, the behavior is sys- tem dependent. rename(s1,s2) fails if unsuccessful, in which case if the file existed previously it is still known by its ori- ginal name. Among the other possible causes of failure would be a file currently open, or a necessity to copy the file's contents to rename it. seek(f,i) seek(f,i) seeks to position i in file f. As with other posi- tions in Icon, a nonpositive value of i can be used to reference a position relative to the end of f. i defaults to 1. seek(f,i) returns f but fails if an error occurs. where(f) where(f) returns the current byte position in the file f. [] Characters and Ordinals char(i) char(i) produces a string containing the character whose internal representation, or ordinal, is the integer i. i must be between 0 and 255 inclusive. If i is out of range, or not con- vertible to integer, a run-time error occurs. ord(s) ord(s) produces an integer between 0 and 255 representing the ordinal, or internal representation, of a character. If s is not convertible to string, or if the string's length is not 1, a run-time error occurs. - 3 - [] Tab Expansion and Insertion detab(s,i1,i2,...,in) detab(s,i1,i2,...,in) replaces each tab character in s by one or more space characters, using tab stops at i1,i2...,in, and then additional tab stops created by repeating the last interval as necessary. The default is detab(s,9). Tab stops must be positive and strictly increasing. There is an implicit tab stop at position 1 to establish the first inter- val. Examples: detab(s) tab stops at 9, 17, 25, 33, ... detab(s,5) tab stops at 5, 9, 13, 17, ... detab(s,8,12) tab stops at 8, 12, 16, 20, ... detab(s,11,18,30,36) tab stops at 11, 18, 30, 36, 42, 48, ... For purposes of tab processing, "\b" has a width of -1, and "\r" and "\n" restart the counting of positions. Other nonprinting characters have zero width, and printing characters have a width of 1. entab(s,i1,i2,...,in) entab(s,i1,i2,...,in) replaces runs of consecutive spaces with tab characters. Tab stops are specified in the same manner as for the detab function. Any existing tab characters in s are preserved, and other nonprinting characters are treated identi- cally with the detab function. The default is entab(s,9). A lone space is never replaced by a tab character; however, a tab character may replace a single space character that is part of a longer run. [] Bit-Wise Functions The following functions operate on the individual bits compos- ing one or two integers. All produce an integer result. iand(i,j) produces the bit-wise and of i and j ior(i,j) produces the bit-wise inclusive or of i and j ixor(i,j) produces the bit-wise exclusive or of i and j icom(i) produces the bit-wise complement (one's complement) of i ishift(i,j) If j is positive, produces i shifted left by j bit positions. If j is negative, produces i shifted right by -j bit positions. If j is zero, produces i. In all cases, vacated bit positions are filled with zeroes. [] Executable Images - 4 - save(s) The function save(s) saves an executable image of an executing Icon program in the file named s. This function presently is implemented only on 4.nbsd UNIX systems. See Section 2.11 for a method of determining if this feature is implemented. save can be called from any point in a program. It accepts a single argument that names the file that is to receive the resulting executable. The named file is created if it does not exist. Any output problems on the file cause save to fail. For lack of anything better, save returns the number of bytes in the data region. When the new executable is run, execution of the Icon program begins in the main procedure. Global and static variables have the value they had when save was called, but all dynamic local variables have the null value. Any initial clauses that have been executed are not re-executed. As usual, arguments present on the command line are passed to the main procedure as a list. Command line input and output redirections are processed normally, but any files that were open are no longer open and attempts to read or write them will fail. When the Icon interpreter starts up, it examines a number of environment variables to determine various operational parame- ters. When a saved executable starts up, the environment vari- ables are not examined; the parameter values recorded in the exe- cutable are used instead. Note that many of the parameter values are dynamic and may have changed considerably from values sup- plied initially. Consider an example: global hello procedure main() initial { hello := "Hello World!" save("hello") | stop("Error saving to 'hello'") exit() } write(hello) end The global variable hello is assigned "Hello World!" and then the interpreter is saved to the file hello. The program then exits. When hello is run, main's initial clause is skipped since it has already been executed. The variable hello has retained its value and the call to write produces the expected greeting. It is possible to call save any number of times during the course of execution. Saving to the same file each time is rather uninteresting, but imagine a complex data structure that passes - 5 - through levels of refinement and then saving out a series of exe- cutables that capture the state of the structure at given times. Saved executables contain the entire data space present at the time of the save and thus can be quite large. Saved executables on the VAX are typically around 250k bytes. Integer Sequences seq(i,j) generates an infinite sequence of integers starting at i with increments of j. An omitted or null-valued argument defaults to 1. For example, seq() generates 1, 2, 3, ... . 2.3 Structures Sets Sets are unordered collections of values and have many of the properties normally associated with sets in the mathematical sense. The function set(a) creates a set that contains the distinct elements of the list a. For example, set(["abc",3]) creates a set with two members, abc and 3. [] The default value for an omitted argument to set is an empty list. Consequently, set() creates an empty set. Sets, like other data aggregates in Icon, need not be homo- geneous - a set may contain members of different types. Sets, like other Icon data aggregates, are represented by pointers to the actual data. Sets can be members of sets, as in s1 := set([1,2,3]) s2 := set([s1,[]]) in which s2 contains two members, one of which is a set of three members and the other of which is an empty list. Any specific value can occur only once in a set. For example, set([1,2,3,3,1]) creates a set with the three members 1, 2, and 3. Set membership - 6 - is determined the same way the equivalence of values is deter- mined in the operation x === y For example, set([[],[]]) creates a set that contains two distinct empty lists. Several set operations are provided. The function member(s,x) succeeds and returns the value of x if x is a member of s, but fails otherwise. Note that member(s1,member(s2,x)) succeeds if x is a member of both s1 and s2. The function insert(s,x) inserts x into the set s and returns the value of s. Note that insert(s,x) is similar to put(a,x) in form. A set may contain (a pointer to) itself: insert(s,s) adds s as an member of itself. The function delete(s,x) deletes the member x from the set s and returns the value of s. The functions insert(s,x) and delete(s,x) always succeed, whether or not x is in s. This allows their use in loops in which failure may occur for other reasons. For example, s := set() while insert(s,read()) builds a set that consists of the (distinct) lines from the stan- dard input file. The operations - 7 - s1 ++ s2 s1 ** s2 s1 -- s2 create the union, intersection, and difference of s1 and s2, respectively. In each case, the result is a new set. The use of these operations on csets is unchanged. There is no automatic type conversion between csets and sets; the result of the operation depends on the types of the arguments. For example, 'aeiou' ++ 'abcde' produces the cset 'abcdeiou', while set([1,2,3]) ++ set([2,3,4]) produces a set that contains 1, 2, 3, and 4. On the other hand, set([1,2,3]) ++ 4 results in Run-time Error 119 (set expected). The functions and operations of Icon that apply to other data aggregates apply to sets as well. For example, if s is a set, *s is the size of s (the number of members in it). Similarly, type(s) produces the string set. The string images of sets are in the same style as for other aggregates, with the size enclosed in parentheses. Therefore, s := set(["abc",3]) write(image(s)) writes set(2). The operation !s generates the members of s, but in no predictable order. Simi- larly, ?s produces a randomly selected member of s. These operations pro- duce values, not variables - it is not possible to assign a value to !s or ?s. - 8 - The function copy(s) produces a new set, distinct from s, but which contains the same members as s. The copy is made in the same fashion as the copy of a list - the members themselves are not copied. The function sort(s) produces a list containing the members of s in sorted order. Sets occur after tables but before records in Icon's collating sequence. Examples Word Counting: The following program lists, in alphabetical order, all the different words that occur in standard input: procedure main() letter := &lcase ++ &ucase words := set() while text := read() do text ? while tab(upto(letter)) do insert(words,tab(many(letter))) every write(!sort(words)) end The Sieve of Eratosthenes: The follow program produces prime numbers, using the classical ``Sieve of Eratosthenes'': procedure main() local limit, s, i limit := 5000 s := set() every insert(s,1 to limit) every member(s,i := 2 to limit) do every delete(s,i + i to limit by i) primes := sort(s) write("There are ",*primes," primes in the first ",limit," integers.") write("The primes are:") every write(right(!primes,*limit + 1)) end - 9 - [] Tables The functions member, insert, and delete apply to tables as well as sets. The function member(t,x) succeeds if x is an entry value (key) in the table t, but fails otherwise. The function insert(t,x,y) inserts the entry value x into table t with the assigned value y. If there already was an entry value x in t, its assigned value is changed. Note that insert has three arguments when used with tables, as compared to two when used with sets. An omitted third argument defaults to the null value. The function delete(t,x) removes the entry value x and its corresponding assigned value from t. If x is not an entry value in t, no operation is performed; delete succeeds in either case. [] Sorting Order for Elements of Lists and Tables A complete ordering is now defined for structures (lists, sets, tables, and records) and csets that appear in a larger structure to be sorted. Different types are still kept separate, but within each structure type, elements are now sorted chrono- logically (in the order they were created). All of the different record types are sorted together. Csets are sorted lexically, in the same fashion as strings. Sorting Options for Tables Two new options are available for sorting tables. These options are specified by the values 3 and 4 as the second argu- ment of sort(t,i). Both of these options produce a single list in which the entry values and assigned values of table elements alternate. A value of 3 for i produces a list in which the entry values are in sorted order, and a value of 4 produces a list in which the assigned values are in sorted order. For example, if the table wcount contains elements whose entry values are words and whose corresponding assigned values are counts, the following code segment writes out a list of the words and their counts, with the words in alphabetical order: a := sort(wcount,3) every i := 1 to *a - 1 by 2 do write(a[i]," : ",a[i + 1]) The main advantage of the new sorting options is that they only produce a single list, rather than a list of lists as pro- duced by the options 1 and 2. The amount of space needed for the single list is proportionally much less than for the list of lists. - 10 - 2.4 Programmer-Defined Control Operations As described in [3], co-expressions can be used to provide programmer-defined control operations. Version 7 provides support for this facility by means of an alternative syntax for procedure invocation in which the arguments are passed in a list of co- expressions. This syntax uses braces in place of parentheses: p{expr1, expr2, ..., exprn} is equivalent to p([create expr1, create expr2, ..., create exprn]) Note that p{} is equivalent to p([]) 2.5 Invocation By String Name A string-valued expression that corresponds to the name of a procedure or operation can be used in place of the procedure or operation in an invocation expression. For example, "image"(x) produces the same call as image(x) and "-"(i,j) is equivalent to i - j In the case of operator symbols with unary and binary forms, the number of arguments determines the operation. Thus "-"(i) is equivalent to -i Since to-by is an operation, despite its reserved-word syntax, it - 11 - is included in this facility with the string name "..." . Thus "..."(1,10,2) is equivalent to 1 to 10 by 2 Similarly, range specifications are represented by ":", so that ":"(s,i,j) is equivalent to s[i:j] The subscripting operation is available with the string name "[]". Thus "[]"(&lcase,3) produces c. Defaults are not provided for omitted or null-valued arguments in this facility. Consequently, "..."(1,10) results in a run-time error when it is evaluated. Arguments to operators invoked by string names are derefer- enced. Consequently, string invocation for assignment operations is ineffective and results in error termination. String names are available for the operations in Icon, but not for control structures. Thus "|"(expr1,expr2) is erroneous. Note that string scanning is a control structure. Field references, of the form expr . fieldname are not operations in the ordinary sense and are not available via string invocation. In addition, conjunction is not available via string invocation, since no operation is actually performed. String names for procedures are available through global iden- tifiers. Note that the names of functions, such as image, are global identifiers. Similarly, any procedure-valued global iden- tifier may be used as the string name of a procedure. Thus in - 12 - global q procedure main() q := p "q"("hi") end procedure p(s) write(s) end the procedure p is invoked via the global identifier q. The function proc(x,i) converts x to a procedure, if possible. If x is procedure-valued, its value is returned unchanged. If the value of x is a string that corresponds to the name of a pro- cedure as described in the preceding section, the corresponding procedure value is returned. The value of i is used to distin- guish between unary and binary operators. For example, proc("*",2) produces the multiplication operator, while proc("*",1) produces the size operator. The default value for i is 1. If x cannot be converted to a procedure, proc(x,i) fails. 2.6 [] String Scanning Scanning environments (&subject and &pos) are now maintained correctly. In particular, they are are now restored when a scan- ning expression is exited via break, next, return, fail, and suspend. This really is a correction of a bug, although the former handling of scanning environments is described as a `feature' in the Icon book. Note also that this change could affect the behavior of existing Icon programs. 2.7 [] Procedures with a Variable Number of Arguments A procedure can be made to accept a variable number of argu- ments by appending [] to the last (or only) parameter in the parameter list. An example is: procedure p(a,b,c[]) suspend a + b + !c end If called as p(1,2,3,4,5), the parameters have the following values: a 1 b 2 c [3,4,5] The last parameter always contains a list. This list consists of the arguments not used by the previous parameters. If the - 13 - previous parameters use up all the arguments, the list is empty. If there are not enough arguments to satisfy the previous parame- ters, the null value will be used for the remaining ones, but the last parameter will still contain the empty list. The following procedure, which models the function write, but keeps a count of the number of arguments written, illustrates a use of this feature: global wcount procedure Write(a[]) initial wcount := 0 wcount +:= *a every writes(!a) write() end 2.8 [] Co-Expressions Co-expression return points are now handled properly, so that co-expressions can be used as coroutines. The image of a co-expression now includes an identifying number. Numbers start with 1 for &main and increase as new co- expressions are created. Co-expression activation and return are now traced along with procedure calls and returns if &trace is nonzero. Co-expression tracing shows the identifying numbers. There is a new keyword, ¤t, the current co-expression. 2.9 [] Input and Output There no longer are any length limitations on the string pro- duced by read() or reads(), nor on the length of the argument to system() or the first argument of open(). Formerly the value returned by write(x1,x2,...,xn) and writes(x1,x2,...,xn) was the last written argument converted to a string. The conversion is no longer performed. For example, the value returned by write(1) is the integer 1. Errors during writing (such as inadequate space on the output device) now cause error termination. The function read(f) reads the last line of the file f, even if that line does not end with a newline (Version 5 discarded such a line). If the file f is open as a pipe, close(f) returns the system - 14 - code resulting from closing f instead of f. Icon no longer performs its own I/O buffering; instead, this function is left to the operating system on which Icon runs. The environment variable NBUFS is no longer supported. 2.10 Errors [] Error Traceback A run-time error now shows a traceback, giving the sequence of procedure calls to the site of the error, followed by a symbolic rendering of the offending expression. For example, suppose the following program is contained in the file max.icn: procedure main() i := max("a",1) end procedure max(i,j) if i > j then i else j end Its execution in Version 7 produces the following output: Run-time error 102 File max.icn; Line 6 numeric expected offending value: "a" Traceback: main() max("a",1) from line 2 in max.icn {"a" > 1} from line 6 in max.icn [] Error Conversion A new keyword, &error, is provided to allow users to convert most run-time errors into expression failure. It behaves like &trace: if it is zero, the default value, errors are handled as usual. If it is non-zero, errors are treated as failure and &error is decremented. There are a few errors that cannot be converted to failure: arithmetic overflow and underflow, stack overflow, and errors during program initialization. When an error is converted to failure in this way, three key- words are set: + &errornumber is the number of the error (e.g., 101). Reference to &errornumber fails if there has not been an error. - 15 - + &errortext is the error message (e.g., integer expected). + &errorvalue is the offending value. Reference to &error- value fails if there is no specific offending value. The function errorclear() removes the indication of the last error. Subsequent references to &errornumber fail until another error occurs. A use of these keywords is illustrated by the following pro- cedure, which could be used to process potential run-time errors that had been converted to failure: procedure ErrorCheck() write("\nError ",&errornumber) write(&errortext) write("offending value: ",image(&errorvalue)) writes("\nDo you want to continue? (n)") if map(read()) == ("y" | "yes") then return else exit() end For example, &error := -1 . . . write(s) | ErrorCheck() could be used to check for an error during writing, while (a := sort(t,3)) | ErrorCheck() could be used to detect failure to sort a table into a list (for lack of adequate storage). The function runerr(i,x) causes program execution to terminate with error number i as if a corresponding run-time error had occurred. If i is the number of a standard run-time error, the corresponding error text is printed; otherwise no error text is printed. The value of x is given as the offending value. If x is omitted, no offending value is printed. This function is provided so that library procedures can be written to terminate in the same fashion as built-in operations. It is advisable to use error numbers for programmer-defined errors that are well outside the range of numbers used by Icon itself. See Appendix B. Error number 500 has the predefined text program malfunction for use with runerr. This number is not used by Icon itself. A call of runerr is subject to conversion to failure like any - 16 - other run-time error. 2.11 [] Implementation Features Since different implementations of Icon may support different features and some installations may chose to provide a subset of the full Version 7, the keyword &features has been provided to provide such information. &features generates the features of the implementation on which the current program is running. For exam- ple, on a 4.nbsd UNIX implementation, every write(&features) produces UNIX co-expressions overflow checking direct execution environment variables error traceback executable images string invocation expandable regions Similarly, a program that uses co-expressions can check for the presence of this feature: if not(&features == "co-expressions") then runerr(-401) 2.12 [] Storage Management Storage is allocated automatically during the execution of an Icon program, and garbage collections are performed automatically to reclaim storage for subsequent reallocation. There are three storage regions: static, string, and block. Only implementations in which regions can be expanded support a static region. See [4] for more information. An Icon programmer normally need not worry about storage management. However, in applications that require a large amount of storage or that must operate in a limited amount of memory, some knowledge of the storage management process may be useful. The keyword &collections generates four values associated with garbage collection: the total number since program initiation, the number triggered by static allocation, the number triggered by string allocation, and the number triggered by block alloca- tion. The keyword ®ions generates the current sizes of the static, string, and block regions. The keyword &storage gen- erates the current amount of space used in the static, string, and block regions. The value given for the static region - 17 - presently is not meaningful. Garbage collection can be forced by the function collect(). 2.13 Version Checking The Icon translator converts a source-language program to an intermediate form, called ucode. The Icon linker converts one or more ucode files to a binary form called icode. The format of Version 7 ucode and icode files is different from that of earlier versions. To avoid the possibility of malfunction due to incom- patible ucode and icode formats, Version 7 checks both ucode and icode files and terminates processing with an error message if the versions are not correct. 2.14 Odds and Ends [] Other Keywords In addition to the new keywords mentioned above, there are three others: + &digits, whose value is the cset '0123456789'. + &line, the current source-code line number. + &file, the current source-code file name. [] External Data Type There is a new data type, external, for use in extensions to Icon that require the allocation of unstructured blocks of data. This data type should not be visible in standard versions of Icon. [] Addition to suspend The suspend control structure now has an optional do clause, analogous to every-do. If a do clause is present in a suspend control structure, its argument is evaluated after the suspend is resumed and before possible suspension with another result. For example, the following expression might be used to count the number of suspensions: suspend expr do count +:= 1 [] Linker Options The option to generate diagnostic (.ux) files during linking has been changed from -D to -L. The amount of space needed to associate source-program line - 18 - numbers with executable code can be set by -Snn. The default value of n is 1000. Similarly, the amount of space needed to associate file names with executable code can be set by -SFn. The default is 10. [] Path to iconx For implementations that support direct execution of icode files, the hardwired path to iconx is overridden by the value of the environment variable ICONX, if it is set. [] File Names During translation and linking, the suffix .u is interpreted as .u1, and no suffix is interpreted as .icn. On MS-DOS, VMS, and the Atari ST the icode file produced by the linker now has the extension icx. For example, icont prog.icn produces an icode file named prog.icx. The extension need not be given when using iconx: iconx prog looks first for prog.icx; failing that it looks for prog. [] Redirection of Error Output The option -e now allows standard error output to be redirected to a file. For example, iconx -e prog.err prog executes prog and sends any error output to the file prog.err. If - is given in place of a file name, error output is redirected to standard output. On systems on which standard output can be redirected to a file, -e - causes both error output and standard output go to that file. For example, iconx -e - prog >prog.out redirects both error output and standard output to prog.out. The use of - with -e is not available on the Atari ST or Macintosh. 3. Obsolete and Changed Features of Version 5 The original implementation of Version 5 supported both a com- piler (iconc) and an interpreter (icont). Version 7 supports only an interpreter. Interpretation is only slightly slower than the execution of compiled code and the interpreter is portable and gets into execution much more quickly than the compiler. - 19 - However, it is not possible to load C functions with the inter- preter as it was with the compiler. A system for personalized interpreters [5] is included with Version 7 for UNIX systems to make it comparatively easy to add new functions and otherwise modify the Icon run-time system. Some run-time environment variables have changed; see Appendix A. Several error messages have been changed. Appendix B con- tains a list of run-time error messages for Version 7. 4. Known Bugs in Version 7 + The text of some translator error messages is incorrect and may not correctly reflect the nature of errors. + The translator does not detect arithmetic overflow in conversion of numeric literals. Very large numeric literals may have incorrect values. + Integer overflow on multiplication and exponentiation may not be detected during execution. Such overflow may occur during type conversion. + Line numbers may be wrong in diagnostic messages related to lines with continued quoted literals. + In some cases, trace messages may show the return of sub- scripted values, such as &null[2], that would be errone- ous if they were dereferenced. + If a long file name for an Icon source-language program is truncated by the operating system, mysterious diagnos- tic messages may occur during linking. + Stack overflow is checked using a heuristic that may not always be effective. + If an expression such as x := create expr is used in a loop, and x is not a global variable, unreferenceable co-expressions are generated by each suc- cessive create operation. These co-expressions are not garbage collected. This problem can be circumvented by making x a global variable or by assigning a value to x before the create operation, as in x := &null x := create expr - 20 - + Stack overflow in a co-expression may not be detected and may cause mysterious program malfunction. 5. Possible Differences Among Version 7 Implementations A few aspects of the implementation of Version 7 are specific to different computer architectures and operating systems. Co- expressions require a context switch that is implemented in assembly language. If this context switch is not implemented, an attempt to activate a co-expression results in error termination. Arithmetic overflow checking also generally requires assembly- language routines and may not be supported on some implementa- tions of Version 7. Some features of Icon, such as opening a pipe for I/O and the system function, are not supported on all operating systems. Acknowledgements In addition to the authors of this report, Dave Gudeman and Bill Mitchell made significant contributions to Version 7 of Icon. References 1. Griswold, Ralph E. and Madge T. Griswold. The Icon Program- ming Language, Prentice-Hall, Inc., Englewood Cliffs, New Jersey. 1983. 2. Griswold, Ralph E. ICONT(1), manual page for UNIX Programmer's Manual, Department of Computer Science, The Univer- sity of Arizona. 1988. 3. Griswold, Ralph E. and Michael Novak. ``Programmer-Defined Control Operations'', The Computer Journal, Vol. 26, No. 2 (May 1983). 4. Griswold, Ralph E. and Madge T. Griswold. The Implementation of the Icon Programming Language, Princeton University Press, Princeton, New Jersey. 1987. 5. Griswold, Ralph E. Personalized Interpreters for Version 7 of Icon, Technical Report TR 88-7, Department of Computer Science, The University of Arizona. 1988. - 21 - Appendix A - Environment Variables There are a number of environment variables that can be set to override the default values for sizes of Icon's storage regions and other run-time parameters. Except for ICONX, NOERRBUF, and ICONCORE, the values assigned to these environment variables must be numbers. Default values for regions vary from system to sys- tem and are given in users manuals. Some implementations also have other environment variables. The environment variables are: ICONX If set, this environment variable is used for the location of iconx used to execute an icode file. TRACE Specifies the initial value of &trace. The default value is zero. NOERRBUF If set, standard error output is not buffered. ICONCORE If set, a core dump is produced in the case of error termination. MSTKSIZE Specifies the size in words of the main inter- preter stack. STRSIZE Specifies the initial size in bytes of the allo- cated string region. HEAPSIZE Specifies the initial size in bytes of the allo- cated block region. STATSIZE Specifies the initial size in bytes of the static region in which co-expressions are allocated. STATINCR Specifies the increment for expanding the static region. The default increment is one-fourth the initial size of the static region. COEXPSIZE Specifies the size in words of co-expression blocks. QLSIZE Specifies the amount of space used for pointers to strings during garbage collection. Used only on implementations with fixed memory regions. An inappropriate setting of an environment variable prevents the program from starting. - 22 - Appendix B - Run-Time Error Messages A list of run-time error numbers and corresponding messages follows. Some implementations have additional run-time errors. 101 integer expected 102 numeric expected 103 string expected 104 cset expected 105 file expected 106 procedure or integer expected 107 record expected 108 list expected 109 string or file expected 110 string or list expected 111 variable expected 112 invalid type to size operation 113 invalid type to random operation 114 invalid type to subscript operation 115 list or table expected 116 invalid type to element generator 117 missing main procedure 118 co-expression expected 119 set expected 120 cset or set expected 121 function not supported 122 set or table expected 201 division by zero 202 remaindering by zero 203 integer overflow 204 real overflow, underflow, or division by zero 205 value out of range 206 negative first operand to real exponentiation 207 invalid field name 208 second and third arguments to map of unequal length 209 invalid second argument to open 210 non-ascending arguments to detab/entab 211 by clause equal to zero 212 attempt to read file not open for reading 213 attempt to write file not open for writing 214 input/output error - 23 - 301 interpreter stack overflow 302 C stack overflow 303 inadequate space for interpreter stack 304 inadequate space in qualifier list 305 inadequate space in static region 306 inadequate space in string region 307 inadequate space in block region 401 co-expressions not implemented 500 program malfunction - 24 -