Chapter 14 The Coq commands

There are three Coq commands:

coqtop: the Coq toplevel (interactive mode);
coqc: the Coq compiler (batch compilation);
coqchk: the Coq checker (validation of compiled libraries).

The options are (basically) the same for the first two commands, and roughly described below. You can also look at the man pages of coqtop and coqc for more details.

14.1 Interactive use (coqtop)

In the interactive mode, also known as the Coq toplevel, the user can develop his theories and proofs step by step. The Coq toplevel is run by the command coqtop.

They are two different binary images of Coq: the byte-code one and the native-code one (if OCaml provides a native-code compiler for your platform, which is supposed in the following). By default, coqtop executes the native-code version; run coqtop.byte to get the byte-code version.

The byte-code toplevel is based on an OCaml toplevel (to allow the dynamic link of tactics). You can switch to the OCaml toplevel with the command Drop., and come back to the Coq toplevel with the command Coqloop.loop();;.

14.2 Batch compilation (coqc)

The coqc command takes a name file as argument. Then it looks for a vernacular file named file.v, and tries to compile it into a file.vo file (See 6.5).

Warning: The name file should be a regular Coq identifier, as defined in Section 1.1. It should contain only letters, digits or underscores (_). For instance, /bar/foo/toto.v is valid, but /bar/foo/to-to.v is invalid.

14.3 Customization at launch time

14.3.1 By resource file

When Coq is launched, with either coqtop or coqc, the resource file $XDG_CONFIG_HOME/coq/coqrc.xxx is loaded, where $XDG_CONFIG_HOME is the configuration directory of the user (by default its home directory /.config and xxx is the version number (e.g. 8.3). If this file is not found, then the file $XDG_CONFIG_HOME/coqrc is searched. You can also specify an arbitrary name for the resource file (see option -init-file below).

This file may contain, for instance, Add LoadPath commands to add directories to the load path of Coq. It is possible to skip the loading of the resource file with the option -q.

14.3.2 By environment variables

Load path can be specified to the Coq system by setting up $COQPATH environment variable. It is a list of directories separated by : (; on windows). Coq will also honor $XDG_DATA_HOME and $XDG_DATA_DIRS (see Section 2.6.3).

Some Coq commands call other Coq commands. In this case, they look for the commands in directory specified by $COQBIN. If this variable is not set, they look for the commands in the executable path.

The $COQ_COLORS environment variable can be used to specify the set of colors used by coqtop to highlight its output. It uses the same syntax as the $LS_COLORS variable from GNU’s ls, that is, a colon-separated list of assignments of the form name=attr1;...;attrn where name is the name of the corresponding highlight tag and attri is an ANSI escape code. The list of highlight tags can be retrieved with the -list-tags command-line option of coqtop.

14.3.3 By command line options

The following command-line options are recognized by the commands coqc and coqtop, unless stated otherwise:

-I directory, -include directory

Add physical path directory to the OCaml loadpath.

See also: Section 2.6.1 and the command Declare ML Module Section 6.5.

-Q directory dirpath

Add physical path directory to the list of directories where Coq looks for a file and bind it to the the logical directory dirpath. The subdirectory structure of directory is recursively available from Coq using absolute names (extending the dirpath prefix) (see Section 2.6.2).

Note that only those subdirectories and files which obey the lexical conventions of what is an ident (see Section 1.1) are taken into account. Conversely, the underlying file systems or operating systems may be more restrictive than Coq. While Linux’s ext4 file system supports any Coq recursive layout (within the limit of 255 bytes per file name), the default on NTFS (Windows) or HFS+ (MacOS X) file systems is on the contrary to disallow two files differing only in the case in the same directory.

14.4 Compiled libraries checker (coqchk)

The coqchk command takes a list of library paths as argument. The corresponding compiled libraries (.vo files) are searched in the path, recursively processing the libraries they depend on. The content of all these libraries is then type-checked. The effect of coqchk is only to return with normal exit code in case of success, and with positive exit code if an error has been found. Error messages are not deemed to help the user understand what is wrong. In the current version, it does not modify the compiled libraries to mark them as successfully checked.

Note that non-logical information is not checked. By logical information, we mean the type and optional body associated to names. It excludes for instance anything related to the concrete syntax of objects (customized syntax rules, association between short and long names), implicit arguments, etc.

This tool can be used for several purposes. One is to check that a compiled library provided by a third-party has not been forged and that loading it cannot introduce inconsistencies.¹ Another point is to get an even higher level of security. Since coqtop can be extended with custom tactics, possibly ill-typed code, it cannot be guaranteed that the produced compiled libraries are correct. coqchk is a standalone verifier, and thus it cannot be tainted by such malicious code.

Command-line options -I, -R, -where and -impredicative-set are supported by coqchk and have the same meaning as for coqtop. Extra options are:

-norec module: Check module but do not check its dependencies.
-admit module: Do not check module and any of its dependencies, unless explicitly required.
-o: At exit, print a summary about the context. List the names of all assumptions and variables (constants without body).
-silent: Do not write progress information in standard output.

Environment variable $COQLIB can be set to override the location of the standard library.

The algorithm for deciding which modules are checked or admitted is the following: assuming that coqchk is called with argument M, option -norec N, and -admit A. Let us write S the set of reflexive transitive dependencies of set S. Then:

Modules C=M\A∪ M∪ N are loaded and type-checked before being added to the context.
And M∪N\ C is the set of modules that are loaded and added to the context without type-checking. Basic integrity checks (checksums) are nonetheless performed.

As a rule of thumb, the -admit can be used to tell that some libraries have already been checked. So coqchk A B can be split in coqchk A && coqchk B -admit A without type-checking any definition twice. Of course, the latter is slightly slower since it makes more disk access. It is also less secure since an attacker might have replaced the compiled library A after it has been read by the first command, but before it has been read by the second command.

1: Ill-formed non-logical information might for instance bind Coq.Init.Logic.True to short name False, so apparently False is inhabited, but using fully qualified names, Coq.Init.Logic.False will always refer to the absurd proposition, what we guarantee is that there is no proof of this latter constant.