TODO 4.46 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
docs
----

  - rename the languages to "Why" and "WhyML". "Why3ML" is a horrible name.

generic
-------

  - let modules register new command-line options. This would deprecate
    the "stop flags" in Debug, and serve, for instance, for --type-only,
    --parse-only, ocaml code extraction, printing of modules, etc.

  - introduce a dedicated buffer-backed formatter for warnings.
    The contents of the buffer would be shown on stdout or in a window
    at selected points of program execution. Demote non-critical errors
    (e.g. unnamed type variables) to warnings.

  - should we create a common [exception Why.Error of exn] to facilitate
    integration of the library? This would require a special [raise] call:
        why_raise e = raise (Why.Error e)

Andrei Paskevich's avatar
Andrei Paskevich committed
22 23 24
  - the drivers cannot deal with theories defined in the .mlw files.
    Otherwise why3 would report an error. Is it acceptable?

25 26 27
WhyML
-----

Andrei Paskevich's avatar
Andrei Paskevich committed
28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
  - introduce logic predicates for program type invariants:
      predicate <type_name>_invariant (result : <type_name>) = ...
    Should this be just the top invariant, as written by the user,
    or should we also include the invariants for the fields?

  - allow to specify type invariants below a type definition,
    provided there are no program declarations in between.
    This allows us to define auxiliary logical functions and
    predicates that depend on the (pure) type and can be used in
    the invariant. However, the parser must know from the start
    that the type has an invariant, what's the best syntax?

  - type invariants are now assumed/asserted at the function call
    boundaries. We can add a binary flag to Ityapp to allow open
    types in function signatures (must be careful with reg_match!).
    The type cast can then play the role of the "close" instruction.
    Do we need it? What's the good syntax for open types?

46 47 48 49 50
  - we check the context type invariants around Eany and after Eabstr.
    It might be strange that Eabstr post-ensures the invariants that
    didn't necessarily hold before its execution. Also, what about
    return/raise invariants, should Eany and Eabstr enforce them?
    (at the moment, they do)
Andrei Paskevich's avatar
Andrei Paskevich committed
51

52
  - currently every unhandled exception has the postcondition "true".
53 54
    "false" would be a poor choice, as it could introduce inconsistency
    in the WPs of the caller. Should we allow unhandled exceptions at all?
Andrei Paskevich's avatar
Andrei Paskevich committed
55

56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77
  - current WP does not respect the lexical scope. In the program

      let r = create 0 in
      let v = !r in
      incr r;
      let () =
        let v = !r in
        ()
      in
      assert { v = 1 }

    the last assert will be proven if the same let_defn [let v = !r]
    and therefore the same pvsymbol v is used in both places (which
    can be done using API). One possible solution is to ensure the
    one-time usage of let_defns and rec_defns in programs.

  - are mutable values worth it? They can only appear as pattern
    variables standing for mutable fields, and allow us to have
    mutable fields in algebraic types. On the other hand, they
    require tedious checks pretty much everywhere in the code,
    and they cannot be translated to OCaml.

78 79
syntax
------
80

81
  - open
Jean-Christophe Filliâtre's avatar
uses  
Jean-Christophe Filliâtre committed
82

83 84 85 86 87 88 89 90
  - infix symbols as constructors, e.g.

       type list 'a = Nil | (::) 'a (list 'a)

  - constants in patterns, e.g.

       match ... with 0 :: r -> ... | ...

91 92
semantics
---------
Andrei Paskevich's avatar
Andrei Paskevich committed
93

Andrei Paskevich's avatar
Andrei Paskevich committed
94 95
  - should split_goal provide a "right-hand side only split"?

96 97 98
  - produce reparsable tasks in Why3 format: how to preserve information about
    the origins of symbols to be able to use drivers after reparsing?

99 100
  - open (et échouer si "open A" et "open B" avec A et B déclarant un symbole
    de même nom)
101

102 103
session
-------
104

105
  - save the output of the prover
106

107
  - escape the string in the xml
108

François Bobot's avatar
François Bobot committed
109
  - the filenames in the location inside a session should be relative
110
    to the session_dir.
111

112 113
  - use the new restore_path for the metas in session?

114 115 116 117
tools
-----

  - the tools should verify that the provers have the same version
118 119 120
    than reported in the configuration
        Andrei: isn't this done?

121
  - Maybe : make something generic for the dialog box with memory.
122 123


124 125 126 127 128 129
OCaml extraction
----------------

  - allow other realizations for arithmetic, such as Zarith or GMP
    (currently this is Num)

130 131 132 133 134
  - avoid conversion to/from int in the for-loop

  - driver
    - %Exit -> Pervasives.Exit

135 136 137 138 139 140
provers
-------

  - PVS: use a better name for PVS theory when printing a task, e.g.
    file_theory_goal. Solution: do that when we have idents with origin
    information (necessary for parsing a task).