curry-repl: A universal interactive environment for Curry compilers

This package contains the implementation of a universal REPL (Read-Eval-Print-Loop) which can be used on top of a Curry compiler. Thus, if the Curry compiler supports some standard options (see below), it can be extended to a simple interactive programming environment.

The specification of the concrete compiler is provided as an element of type REPL.Compiler.CCDescription. The entry point of the REPL is the operation mainREPL defined in module REPL.Main.

The directory examples contains example specifications for various Curry systems.

The directory scripts contains a simple shell script to invoke a Curry REPL with appropriate path settings. This script needs to be adapted to the actual Curry compiler.

Requirements for a Curry Compiler used by a REPL

Basically, the REPL translates each expression to be evaluated into a program with a main operation. This program is compiled and then executed. In order to implement this kind of REPL, it should be possible to invoke the Curry compiler curryc as follows:

curryc [-vn] [-iDIR] ... [-iDIR] [--parse-options="PO"] [-c|--compile] MOD

where the possible arguments are:

• -vn: verbosity with 0 <= n <= 4
• -iDIR: use DIR to search for imported modules
• -c: compile only
• PO: additional options passed to the Curry front end
• MOD: the module to be compiled containing an operation main

This command should compile module MOD (and, if necessary, all imported modules) and create an executable MOD which executes the operation main in this module. MOD might be a hierarchical module name, e.g., Dir.Mod. In this case, the executable Dir.Mod is generated relative to the source directory, e.g., if the source code of the module is stored in Dir/Mod.curry, the executable is stored in the file Dir.Mod.

If the option -c or --compile is provided, the executable is not generated. This might be reasonable to compile all modules in order to check for errors and speed up later compilations without re-compiling all imports.

If the module name is prefixed by a path, e.g., dir1/dir2/Mod, then the REPL changes into the directory of the path (dir1/dir2) and compile the main module there.

Further options:

Options beyond this general usage might depend on the compiler. For instance, each compiler might implement a different set of search strategies. Here are some examples of options and their values:

• interactive (yes/no): should the user be asked to print the next result?
• first (yes/no): print only one result and stop?
• results (int n): number of results to be printed (n=0: infinite)
• time (yes/no): print execution time?
• strategy: search strategy to be used, e.g., dfs/bfs/ids/fs
• debug (yes/no): interactive debugging
• trace (yes/no): trace evaluation
• printdepth (int d): print resulting terms up to depth d (d=0: no limit)

The actual options are specified by data of type CCOption (see module REPL.Compiler).

RC file:

The REPL reads on startup a compiler-specific configuration file which contains definitions for some options and commands, like a search path for additional libraries, commands to show or edit source files, etc. A template for such a configuration file is contained as curryrc.default in this package. The REPL assumes that the home directory of the Curry compiler contains a configuration file CYCrc.default if CYC is the name of the compiler. If there is none, the configuration file curryrc.default of this package is used. On startup, the REPL copies this file (without the suffic .default) into the user’s home directory, if there is not already one, otherwise the file is updated if the default file contains new fields.

Requirements for Curry Compilers used by CPM

The Curry REPL is the basic interface for an interactive Curry system. Thus, many Curry systems implement the commands and options provided by this universal REPL. This property is exploited by the Curry Package Manager (CPM) in order to support the use of packages for various Curry systems. The actual requirements of CPM are described in the following.

In order to use a Curry system together with the Curry Package Manager CPM, the Curry system has to support some options so that CPM can interact with it. The universal Curry REPL implements these options provided that the Curry compiler itself provides options about its version. If curryc is the executable of the compiler, the following options must exist:

• curryc --compiler-name: Shows that name of the compiler (which occurs in compiler dependencies of package specifications) and quits.
• curryc --numeric-version: Shows the compiler version quits.
• curryc --base-version: Shows the version of the base libraries implemented by the compiler and quits.

Note that the universal Curry REPL provides the same options, but they are implemented by passing them to the compiler.

These options can also be combined. In this case the information is shown in subsequent lines, as shown in this example for PAKCS:

> pakcs --compiler-name --numeric-version --base-version
pakcs
3.3.0
3.0.0

In order to support the use of different Curry compilers in parallel, the compiler should store its generated intermediate files in the directory

.curry/<compiler-name>-<numeric-version>

relative to the directory of the source file. If the source file is a hierarchical module, the same hierarchy is used relative to .curry. For instance, if the PAKCS compiler version 3.3.0 is used to compile a hierarchical module Dir.Mod stored in the file

/home/curry/base/Dir/Mod.curry

the Prolog target file is stored in

/home/curry/base/.curry/pakcs-3.3.0/Dir/Mod.pl

Running programs

In order to use a Curry system (and its REPL) conveniently, the Curry system should query CPM (by cypm deps -p) to get the value of CURRYPATH to load modules at startup time. This is usually done by a separate shell script which invokes the actual executable of the REPL.

When CPM starts a Curry system (via the command cypm curry), it sets the environment variable CURRYPATH to the load path of all included packages and passes the option --nocypm to the Curry system. This option should have the effect that CPM is not invoked at startup time (to avoid a cyclic invocation of both systems). The implementation of the universal Curry REPL just ignores the option --nocypm (as mentioned above, the querying of CPM is usually done by a separate shell script which invokes the REPL).

Installing executables

If CPM installs an executable (by the command cypm install), it passes the following options to the Curry system (where curry denotes the main executable of the Curry system, i.e., the REPL):

> curry :set v0 :load MAINMOD :save :quit

(where v0 is replaced by v1 in debug mode). The execution of this command should install the executable MAINMOD. Hence, these options are also implemented by the universal Curry REPL.