6 Reader Helpers

6.1 Raising exn:fail:read

Source-location information is added to the error message using the last five arguments and the extra-srclocs (if the error-print-source-location parameter is set to #t). The source argument is an arbitrary value naming the source location—usually a file path string. Each of the line, pos arguments is #f or a positive exact integer representing the location within source-name (as much as known), col is a non-negative exact integer for the source column (if known), and span is #f or a non-negative exact integer for an item range starting from the indicated position.

The usual location values should point at the beginning of whatever it is you were reading, and the span usually goes to the point the error was discovered.

6.2 Module Reader

See also Defining new #lang Languages in The Racket Guide.

The syntax/module-reader library provides support for defining #lang readers. It is normally used as a module language, though it may also be required to get make-meta-reader. It provides all of the bindings of racket/base other than #%module-begin.

(module reader syntax/module-reader

module-path)

creates a reader such that a module source

#lang something

....

is read as

(module name-id module-path

(#%module-begin ....))

where name-id is derived from the source input port’s name in the same way as for #lang s-exp.

Keyword-based reader-options allow further customization, as listed below. Additional forms are as in the body of racket/base module; they can import bindings and define identifiers used by the reader-options.

• #:read and #:read-syntax (both or neither must be supplied) specify alternate readers for parsing the module body—replacements read and read-syntax, respectively. Normally, the replacements for read and read-syntax are applied repeatedly to the module source until eof is produced, but see also #:whole-body-readers?.

  Unless #:whole-body-readers? specifies a true value, the repeated use of read or read-syntax is parameterized to set read-accept-lang to #f, which disables nested uses of #lang.

  See also #:wrapper1 and #:wrapper2, which support simple parameterization of readers rather than wholesale replacement.

• #:whole-body-readers? specified as true indicates that the #:read and #:read-syntax functions each produce a list of S-expressions or syntax objects for the module content, so that each is applied just once to the input stream.

  If the resulting list contains a single form that starts with the symbol '#%module-begin (or a syntax object whose datum is that symbol), then the first item is used as the module body; otherwise, a '#%module-begin (symbol or identifier) is added to the beginning of the list to form the module body.

• #:wrapper1 specifies a function that controls the dynamic context in which the read and read-syntax functions are called. A #:wrapper1-specified function must accept a thunk, and it normally calls the thunk to produce a result while parameterizing the call. Optionally, a #:wrapper1-specified function can accept a boolean that indicates whether it is used in read (#f) or read-syntax (#t) mode.

  For example, a language like racket/base but with case-insensitive reading of symbols and identifiers can be implemented as

  (module reader syntax/module-reader

  racket/base

  #:wrapper1 (lambda (t)

  (parameterize ([read-case-sensitive #f])

  (t))))

  Using a readtable, you can implement languages that are extensions of plain S-expressions.

• #:wrapper2 is like #:wrapper1, but a #:wrapper2-specified function receives the input port to be read, and the function that it receives accepts an input port (usually, but not necessarily the same input port). A #:wrapper2-specified function can optionally accept an boolean that indicates whether it is used in read (#f) or read-syntax (#t) mode.

• #:info specifies an implementation of reflective information that is used by external tools to manipulate the source of modules in the language something. For example, DrRacket uses information from #:info to determine the style of syntax coloring that it should use for editing a module’s source.

  The #:info specification should be a function of three arguments: a symbol indicating the kind of information requested (as defined by external tools), a default value that normally should be returned if the symbol is not recognized, and a default-filtering function that takes the first two arguments and returns a result.

  The expression after #:info is placed into a context where language-module and language-data are bound. The language-module identifier is bound to the module-path that is used for the read module’s language as written directly or as determined through #:language. The language-data identifier is bound to the second result from #:language, or #f by default.

  The default-filtering function passed to the #:info function is intended to provide support for information that syntax/module-reader can provide automatically. Currently, it recognizes only the 'module-language key, for which it returns language-module; it returns the given default value for any other key.

  In the case of the DrRacket syntax-coloring example, DrRacket supplies 'color-lexer as the symbol argument, and it supplies #f as the default. The default-filtering argument (i.e., the third argument to the #:info function) currently just returns the default for 'color-lexer.

• #:language-info specifies an implementation of reflective information that is used by external tools to manipulate the module in the language something in its expanded, compiled, or declared form (as opposed to source). For example, when Racket starts a program, it uses information attached to the main module to initialize the run-time environment.

  Submodules are normally a better way to implement reflective information, instead of #:language-info. For example, when Racket starts a program, it also checks for a configure-runtime submodule of the main module to initialize the run-time environment. The #:language-info mechanism pre-dates submodules.

  Since the expanded/compiled/declared form exists at a different time than when the source is read, a #:language-info specification is a vector that indicates an implementation of the reflective information, instead of a direct implementation as a function like #:info. The first element of the vector is a module path, the second is a symbol corresponding to a function exported from the module, and the last element is a value to be passed to the function. The last value in the vector must be one that can be written with write and read back with read. When the exported function indicated by the first two vector elements is called with the value from the last vector element, the result should be a function or two arguments: a symbol and a default value. The symbol and default value are used as for the #:info function (but without an extra default-filtering function).

  The value specified by #:language-info is attached to the module form that is parsed from source through the 'module-language syntax property. See module for more information.

  The expression after #:language-info is placed into a context where language-module are language-data are bound, the same as for #:info.

  In the case of the Racket run-time configuration example, Racket uses the #:language-info vector to obtain a function, and then it passes 'configure-runtime to the function to obtain information about configuring the runtime environment. See also Language Run-Time Configuration.

• #:language allows the language of the read module to be computed dynamically and based on the program source, instead of using a constant module-path. (Either #:language or module-path must be provided, but not both.)

  This value of the #:language option can be either a module path (possibly as a syntax object) that is used as a module language, or it can be a procedure. If it is a procedure it can accept either

  • 0 arguments;

  • 1 argument: an input port; or

  • 5 arguments: an input port, a syntax object whose datum is a module path for the enclosing module as it was referenced through #lang or #reader, a starting line number (positive exact integer) or #f, a column number (non-negative exact integer) or #f, and a position number (positive exact integer) or #f.

  The result can be either

  • a single value, which is a module path or a syntax object whose datum is a module path, to be used like module-path; or

  • two values, where the first is like a single-value result and the second can be any value.

  The second result, which defaults to #f if only a single result is produced, is made available to the #:info and #:module-info functions through the language-data binding. For example, it can be a specification derived from the input stream that changes the module’s reflective information (such as the syntax-coloring mode or the output-printing styles).

As another example, the following reader defines a “language” that ignores the contents of the file, and simply reads files as if they were empty:

(module ignored syntax/module-reader

racket/base

#:wrapper1 (lambda (t) (t) '()))

Note that the wrapper still performs the read, otherwise the module loader would complain about extra expressions.

As a more useful example, the following module language is similar to at-exp, where the first datum in the file determines the actual language (which means that the library specification is effectively ignored):

(module reader syntax/module-reader

-ignored-

#:wrapper2

(lambda (in rd stx?)

(let* ([lang (read in)]

[mod  (parameterize ([current-readtable

(make-at-readtable)])

(rd in))]

[mod  (if stx? mod (datum->syntax #f mod))]

[r (syntax-case mod ()

[(module name lang* . body)

(with-syntax ([lang (datum->syntax

#'lang* lang #'lang*)])

(syntax/loc mod (module name lang . body)))])])

(if stx? r (syntax->datum r))))

(require scribble/reader))

The ability to change the language position in the resulting module expression can be useful in cases such as the above, where the base language module is chosen based on the input. To make this more convenient, you can omit the module-path and instead specify it via a #:language expression. This expression can evaluate to a datum or syntax object that is used as a language, or it can evaluate to a thunk. In the latter case, the thunk is invoked to obtain such a datum before reading the module body begins, in a dynamic extent where current-input-port is the source input. A syntax object is converted using syntax->datum when a datum is needed (for read instead of read-syntax). Using #:language, the last example above can be written more concisely:

(module reader syntax/module-reader

#:language read

#:wrapper2 (lambda (in rd stx?)

(parameterize ([current-readtable

(make-at-readtable)])

(rd in)))

(require scribble/reader))

For such cases, however, the alternative reader constructor make-meta-reader implements a might tightly controlled reading of the module language.

The at-exp, reader, and planet languages are implemented using this function.

The generated functions expect a target language description in the input stream that is provided to read-spec. The default read-spec extracts a non-empty sequence of bytes after one or more space and tab bytes, stopping at the first whitespace byte or end-of-file (whichever is first), and it produces either such a byte string or #f. If read-spec produces #f, a reader exception is raised, and path-desc-str is used as a description of the expected language form in the error message.

The reader language supplies read for read-spec. The at-exp and planet languages use the default read-spec.

The result of read-spec is converted to a module path using module-path-parser. If module-path-parser produces a vector of module paths, they are tried in order using module-declared?. If module-path-parser produces #f, a reader exception is raised in the same way as when read-spec produces a #f. The planet languages supply a module-path-parser that converts a byte string to a module path.

If loading the module produced by module-path-parser succeeds, then the loaded module’s read, read-syntax, or get-info export is passed to convert-read, convert-read-syntax, or convert-get-info, respectively.

The at-exp language supplies convert-read and convert-read-syntax to add @-expression support to the current readtable before chaining to the given procedures.

The procedures generated by make-meta-reader are not meant for use with the syntax/module-reader language; they are meant to be exported directly.

Repeatedly calls read on in until an end of file, collecting the results in order into lst, and derives a name-id from (object-name in) in the same way as #lang s-exp. The last five arguments are used to construct the syntax object for the language position of the module. The result is roughly

`(module ,name-id ,mod-path ,@lst)