CoqStyle - coq/coq GitHub Wiki
Here we will present ideas of what constitutes a well written Coq source file. We discuss the general issues that affect the shape of the Coq scripts and consequently their legibility and re-usability. Yes, we are interested in reading and editing ASCII Coq sources!
This is intended to be a collection of suggestions for good style and good practice, acknowledging that usually more than one way is possible.
For the discussion of issues related to the content of specific formalisations and general (type theoretic) tips and traps to avoid please use TipsAndTricks.
- Following the shape of the files in the standard library of Coq, start the Coq files with a comment of the form
(************************************************************************)
(* Copyright <YEAR> <AUTHOR> *)
(* <LICENSE> *)
(************************************************************************)
This will credit the author(s) and clarifies any license issues, which is necessary for re-usability of your code.
- Use coqdoc-style comments graciously, to clarify the structure of your formalisation. (However, try not to use them inside proofs. Instead use ordinary
(* *)
comments. This is because proofs are less likely to be processed via coqdoc and usually they are suppressed.) - Use blank lines consistently (I.e. use the same number of blank lines between all lemmas in a section).
- End the file with a newline character.
- Start the proofs using a line containing
Proof.
- When a tactic generates more than one subgoal use indentation for tactics that are used for solving each subgoal.
- Use comments when using complicated or lengthy proof segments on subgoals of
case
,induction
,destruct
and their friends. - Use blank lines to separate different segments of the proof.
Lemma foo:
Proof.
<body of tactics before case>.
case (compare_with_zero X).
(* 0 < X *)
<body of tactics>...
(* X < 0 *)
<body of tactics>...
(* X = 0 *)
<body of tactics>...
<body of tactics after case>.
Qed.
An alternate possible layout is to use indentation to tell exactly how many subgoals remain: n indentation for n subgoals (variant: (n-1)). This is especially useful during development when you replay proofs after you have modified a definition above: you can easily find where your proof script behaves differently.
The previous example would look much less nice than with the former style; on the contrary this latter style fits better with a forward proof-style, using assert-like tactics.
Example:
Lemma bar:
Proof.
assert (H: <some property>).
rewrite foobar.
intuition.
assert (H2: <something else>).
simpl.
omega.
assert (H3: ...).
auto.
<body of tactics>
Qed.
- It is good practice to use a uniform naming convention for inductive data-types, their constructors, the inverse of the constructors (provable by
inversion
) and the strong elimination rules (generated byScheme
). - For lemmas try to follow a uniform naming convention based on the subject of the formalisation. If in doubt use the naming pattern used for similar lemmas that is used in the Coq standard library.
- It is usually a good idea to explicitly give names to all the hypothesis that are introduced by Coq. This makes reuse of the code easier. It is easy to achieve this when using
intros
, while fordestruct
,induction
and similar tactics, this means that you have to use theas
keyword to explicitly give names. - Orthogonal to the above, at some points it is better to use underscore
_
to immediately clear the variables that are not useful.
- When extracting to Haskell, note that Haskell module names should start with an upper-case letter. It is best practice to use the whole Coq generated code as a Haskell module and write your own pretty printing interface for it (in fact this is the default behaviour of the extraction mechanism). Instead of simply capitalising the first letter, I usually use something like following
Extraction Language Haskell.
Extraction "Xfoo.hs" foo.
to extract a Coq constant
foo
into Haskell moduleXfoo.hs
. This way it is emphasised that this Haskell file is a result of Coq extraction.