ProofMode.md 16.9 KB
Newer Older
1 2 3
Tactic overview
===============

4 5
Many of the tactics below apply to more goals than described in this document
since the behavior of these tactics can be tuned via instances of the type
6 7 8
classes in the file [proofmode/classes](proofmode/classes.v). Most notably, many
of the tactics can be applied when the connective to be introduced or to be eliminated
appears under a later, an update modality, or in the conclusion of a
9
weakest precondition.
10

11 12 13 14 15 16 17 18 19 20 21 22
Starting and stopping the proof mode
------------------------------------

- `iStartProof PROP` : start the proof mode by turning a Coq goal into a proof
  mode entailment. This tactic is performed implicitly by all proof mode tactics
  described in this file, and thus should generally not be used by hand. The
  optional argument `PROP` can be used to explicitly specify which BI logic
  `PROP : bi` should be used. This is useful to drop down in a layered logic,
  e.g. to drop down from `monPred PROP` to `PROP`.
- `iStopProof` to turn the proof mode entailment into an ordinary Coq goal
  `big star of context ⊢ proof mode goal`.

23 24 25 26 27
Applying hypotheses and lemmas
------------------------------

- `iExact "H"`  : finish the goal if the conclusion matches the hypothesis `H`
- `iAssumption` : finish the goal if the conclusion matches any hypothesis
28
- `iApply pm_trm` : match the conclusion of the current goal against the
29 30
  conclusion of `pm_trm` and generates goals for the premises of `pm_trm`. See
  proof mode terms below.
Ralf Jung's avatar
Ralf Jung committed
31
  If the applied term has more premises than given specialization patterns, the
32
  pattern is extended with `[] ... []`.  As a consequence, all unused spatial
Ralf Jung's avatar
Ralf Jung committed
33
  hypotheses move to the last premise.
34 35 36 37

Context management
------------------

38
- `iIntros (x1 ... xn) "ipat1 ... ipatn"` : introduce universal quantifiers
39 40
  using Coq introduction patterns `x1 ... xn` and implications/wands using proof
  mode introduction patterns `ipat1 ... ipatn`.
41 42 43 44
- `iClear (x1 ... xn) "selpat"` : clear the hypotheses given by the selection
  pattern `selpat` and the Coq level hypotheses/variables `x1 ... xn`.
- `iRevert (x1 ... xn) "selpat"` : revert the hypotheses given by the selection
  pattern `selpat` into wands, and the Coq level hypotheses/variables
45 46
  `x1 ... xn` into universal quantifiers. Intuitionistic hypotheses are wrapped
  into the intuitionistic modality.
47
- `iRename "H1" into "H2"` : rename the hypothesis `H1` into `H2`.
48 49
- `iSpecialize pm_trm` : instantiate universal quantifiers and eliminate
  implications/wands of a hypothesis `pm_trm`. See proof mode terms below.
50
- `iSpecialize pm_trm as #` : instantiate universal quantifiers and eliminate
51 52 53
  implications/wands of a hypothesis `pm_trm` whose conclusion is persistent.
  All hypotheses can be used for proving the premises of `pm_trm`, as well as
  for the resulting main goal.
54 55
- `iPoseProof pm_trm as (x1 ... xn) "ipat"` : put `pm_trm` into the context and
  eliminates it. This tactic is essentially the same as `iDestruct` with the
56
  difference that when `pm_trm` is a non-universally quantified intuitionistic
57
  hypothesis, it will not throw the hypothesis away.
Robbert Krebbers's avatar
Robbert Krebbers committed
58 59 60 61 62
- `iAssert P with "spat" as "ipat"` : generates a new subgoal `P` and adds the
  hypothesis `P` to the current goal. The specialization pattern `spat`
  specifies which hypotheses will be consumed by proving `P`. The introduction
  pattern `ipat` specifies how to eliminate `P`.
  In case all branches of `ipat` start with a `#` (which causes `P` to be moved
63 64 65
  to the intuitionistic context) or with an `%` (which causes `P` to be moved to
  the pure Coq context), then one can use all hypotheses for proving `P` as well
  as for proving the current goal.
Robbert Krebbers's avatar
Robbert Krebbers committed
66 67 68
- `iAssert P as %cpat` : assert `P` and eliminate it using the Coq introduction
  pattern `cpat`. All hypotheses can be used for proving `P` as well as for
  proving the current goal.
69 70 71 72 73

Introduction of logical connectives
-----------------------------------

- `iPureIntro` : turn a pure goal into a Coq goal. This tactic works for goals
74
  of the shape `⌜φ⌝`, `a ≡ b` on discrete OFEs, and `✓ a` on discrete cameras.
75 76 77 78 79 80 81 82

- `iLeft` : left introduction of disjunction.
- `iRight` : right introduction of disjunction.

- `iSplit` : introduction of a conjunction, or separating conjunction provided
  one of the operands is persistent.
- `iSplitL "H1 ... Hn"` : introduction of a separating conjunction. The
  hypotheses `H1 ... Hn` are used for the left conjunct, and the remaining ones
83 84
  for the right conjunct. Intuitionistic hypotheses are ignored, since these do
  not need to be split.
85 86 87 88 89 90 91
- `iSplitR "H0 ... Hn"` : symmetric version of the above.
- `iExist t1, .., tn` : introduction of an existential quantifier.

Elimination of logical connectives
----------------------------------

- `iExFalso` : Ex falso sequitur quod libet.
Robbert Krebbers's avatar
Robbert Krebbers committed
92 93 94 95 96
- `iDestruct pm_trm as (x1 ... xn) "ipat"` : elimination of a series of
  existential quantifiers using Coq introduction patterns `x1 ... xn`, and
  elimination of an object level connective using the proof mode introduction
  pattern `ipat`.
  In case all branches of `ipat` start with a `#` (which causes the hypothesis
97
  to be moved to the intuitionistic context) or with an `%` (which causes the
Robbert Krebbers's avatar
Robbert Krebbers committed
98 99 100 101
  hypothesis to be moved to the pure Coq context), then one can use all
  hypotheses for proving the premises of `pm_trm`, as well as for proving the
  resulting goal. Note that in this case the hypotheses still need to be
  subdivided among the spatial premises.
102
- `iDestruct pm_trm as %cpat` : elimination of a pure hypothesis using the Coq
103 104 105
  introduction pattern `cpat`. When using this tactic, all hypotheses can be
  used for proving the premises of `pm_trm`, as well as for proving the
  resulting goal.
106

107
Separation logic-specific tactics
108 109
---------------------------------

110 111 112
- `iFrame (t1 .. tn) "selpat"` : cancel the Coq terms (or Coq hypotheses)
  `t1 ... tn` and Iris hypotheses given by `selpat` in the goal. The constructs
  of the selection pattern have the following meaning:
113 114

  + `%` : repeatedly frame hypotheses from the Coq context.
115
  + `#` : repeatedly frame hypotheses from the intuitionistic context.
116 117
  + `∗` : frame as much of the spatial context as possible. (N.B: this
          is the unicode symbol `∗`, not the regular asterisk `*`.)
118 119

  Notice that framing spatial hypotheses makes them disappear, but framing Coq
120
  or intuitionistic hypotheses does not make them disappear.
121

122
  This tactic solves the goal if everything in the conclusion has been
123
  framed.
124 125
- `iCombine "H1" "H2" as "pat"` : combines `H1 : P1` and `H2 : P2` into
  `H: P1 ∗ P2`, then calls `iDestruct H as pat` on the combined hypothesis.
Robbert Krebbers's avatar
Robbert Krebbers committed
126 127 128
- `iAccu` : solves a goal that is an evar by instantiating it with a all
  hypotheses from the spatial context joined together with a separating
  conjunction (or `emp` in case the spatial context is empty).
129

130 131 132
Modalities
----------

133 134 135
- `iModIntro mod` : introduction of a modality. The type class `FromModal` is
  used to specify which modalities this tactic should introduce. Instances of
  that type class include: later, except 0, basic update and fancy update,
136 137 138
  intuitionistically, persistently, affinely, plainly, absorbingly, objectively,
  and subjectively. The optional argument `mod` can be used to specify what
  modality to introduce in case of ambiguity, e.g. `⎡|==> P⎤`.
139
- `iAlways` : a deprecated alias of `iModIntro`.
140 141
- `iNext n` : an alias of `iModIntro (▷^n P)`.
- `iNext` : an alias of `iModIntro (▷^1 P)`.
142 143 144 145
- `iMod pm_trm as (x1 ... xn) "ipat"` : eliminate a modality `pm_trm` that is
  an instance of the `ElimModal` type class. Instances include: later, except 0,
  basic update and fancy update.

146 147
Induction
---------
148

149 150 151 152
- `iLöb as "IH" forall (x1 ... xn) "selpat"` : perform Löb induction by
  generating a hypothesis `IH : ▷ goal`. The tactic generalizes over the Coq
  level variables `x1 ... xn`, the hypotheses given by the selection pattern
  `selpat`, and the spatial context.
153 154
- `iInduction x as cpat "IH" forall (x1 ... xn) "selpat"` : perform induction on
  the Coq term `x`. The Coq introduction pattern is used to name the introduced
155 156 157 158
  variables. The induction hypotheses are inserted into the intuitionistic
  context and given fresh names prefixed `IH`. The tactic generalizes over the
  Coq level variables `x1 ... xn`, the hypotheses given by the selection pattern
  `selpat`, and the spatial context.
Robbert Krebbers's avatar
Robbert Krebbers committed
159

Robbert Krebbers's avatar
Robbert Krebbers committed
160 161 162 163 164
Rewriting / simplification
--------------------------

- `iRewrite pm_trm` / `iRewrite pm_trm in "H"` : rewrite using an internal
  equality in the proof mode goal / hypothesis `H`.
165 166
- `iRewrite -pm_trm` / `iRewrite -pm_trm in "H"` : rewrite in reverse direction
  using an internal equality in the proof mode goal / hypothesis `H`.
167 168 169 170 171 172 173 174 175 176 177 178 179
- `iEval (tac)` / `iEval (tac) in "selpat"` : performs a tactic `tac`
  on the proof mode goal / hypotheses given by the selection pattern
  `selpat`. Using `%` as part of the selection pattern is unsupported.
  The tactic `tac` should be a reduction or rewriting tactic like
  `simpl`, `cbv`, `lazy`, `rewrite` or `setoid_rewrite`. The `iEval`
  tactic is implemented by running `tac` on `?evar ⊢ P` / `P ⊢ ?evar`
  where `P` is the proof goal / a hypothesis given by `selpat`. After
  running `tac`, `?evar` is unified with the resulting `P`, which in
  turn becomes the new proof mode goal / a hypothesis given by
  `selpat`. Note that parentheses around `tac` are needed.
- `iSimpl` / `iSimpl in "selpat"` : performs `simpl` on the proof mode
  goal / hypotheses given by the selection pattern `selpat`. This is a
  shorthand for `iEval (simpl)`.
180 181 182 183 184


Iris
----

185
- `iInv S with "selpat" as (x1 ... xn) "ipat" "Hclose"` : where `S` is either
186 187
   a namespace `N` or an identifier `H`. Open the invariant indicated by `S`.
   The selection pattern `selpat` is used for any auxiliary assertions needed to
188 189
   open the invariant (e.g. for cancelable or non-atomic invariants). The update
   for closing the invariant is put in a hypothesis named `Hclose`.
190 191 192 193 194 195 196 197

Miscellaneous
-------------

- The tactic `done` is extended so that it also performs `iAssumption` and
  introduces pure connectives.
- The proof mode adds hints to the core `eauto` database so that `eauto`
  automatically introduces: conjunctions and disjunctions, universal and
198 199
  existential quantifiers, implications and wand, plainness, persistence, later
  and update modalities, and pure connectives.
200

201 202 203 204 205 206 207 208 209
Selection patterns
==================

Selection patterns are used to select hypotheses in the tactics `iRevert`,
`iClear`, `iFrame`, `iLöb` and `iInduction`. The proof mode supports the
following _selection patterns_:

- `H` : select the hypothesis named `H`.
- `%` : select the entire pure/Coq context.
210
- `#` : select the entire intuitionistic context.
211 212
- `∗` : select the entire spatial context. (N.B: this
        is the unicode symbol `∗`, not the regular asterisk `*`.)
213

214 215 216 217 218
Introduction patterns
=====================

Introduction patterns are used to perform introductions and eliminations of
multiple connectives on the fly. The proof mode supports the following
219
_introduction patterns_:
220

221
- `H` : create a hypothesis named `H`.
222 223 224
- `?` : create an anonymous hypothesis.
- `_` : remove the hypothesis.
- `$` : frame the hypothesis in the goal.
225 226 227 228 229 230
- `[ipat1 ipat2]` : (separating) conjunction elimination. In order to eliminate
  conjunctions `P ∧ Q` in the spatial context, one of the following conditions
  should hold:
  + Either the proposition `P` or `Q` should be persistent.
  + Either `ipat1` or `ipat2` should be `_`, which results in one of the
    conjuncts to be thrown away.
231 232
- `(pat1 & pat2 & ... & patn)` : syntactic sugar for `[pat1 [pat2 .. patn ..]]`
  to eliminate nested (separating) conjunctions.
233
- `[ipat1|ipat2]` : disjunction elimination.
234
- `[]` : false elimination.
235
- `%` : move the hypothesis to the pure Coq context (anonymously).
236
- `->` and `<-` : rewrite using a pure Coq equality
237
- `# ipat` : move the hypothesis into the intuitionistic context.
238
- `> ipat` : eliminate a modality (if the goal permits).
239 240 241 242

Apart from this, there are the following introduction patterns that can only
appear at the top level:

243 244 245
- `{selpat}` : clear the hypotheses given by the selection pattern `selpat`.
  Items of the selection pattern can be prefixed with `$`, which cause them to
  be framed instead of cleared.
246
- `!%` : introduce a pure goal (and leave the proof mode).
247 248
- `!>` : introduce a modality by calling `iModIntro`.
- `!#` : introduce a modality by calling `iModIntro` (deprecated).
249
- `/=` : perform `simpl`.
250 251
- `//` : perform `try done` on all goals.
- `//=` : syntactic sugar for `/= //`
252 253 254 255
- `*` : introduce all universal quantifiers.
- `**` : introduce all universal quantifiers, as well as all arrows and wands.

For example, given:
256

257 258 259
        ∀ x, <affine> ⌜ x = 0 ⌝ ⊢
          □ (P → False ∨ □ (Q ∧ ▷ R) -∗
          P ∗ ▷ (R ∗ Q ∧ ⌜ x = pred 2 ⌝)).
260

261
You can write
262

263
        iIntros (x Hx) "!# $ [[] | #[HQ HR]] /= !>".
264 265

which results in:
266

267 268 269
        x : nat
        H : x = 0
        ______________________________________(1/1)
Robbert Krebbers's avatar
Robbert Krebbers committed
270
        "HQ" : Q
271 272
        "HR" : R
        --------------------------------------□
273
        R ∗ Q ∧ x = 1
274 275


276 277
Specialization patterns
=======================
278

279
Since we are reasoning in a spatial logic, when eliminating a lemma or
280
hypothesis of type ``P_0 -∗ ... -∗ P_n -∗ R``, one has to specify how the
281
hypotheses are split between the premises. The proof mode supports the following
282
_specification patterns_ to express splitting of hypotheses:
283

284 285
- `H` : use the hypothesis `H` (it should match the premise exactly). If `H` is
  spatial, it will be consumed.
286

287 288 289 290 291
- `(H spat1 .. spatn)` : first recursively specialize the hypothesis `H` using
  the specialization patterns `spat1 .. spatn`, and finally use the result of
  the specialization of `H` (it should match the premise exactly). If `H` is
  spatial, it will be consumed.

292
- `[H1 .. Hn]` and `[H1 .. Hn //]` : generate a goal for the premise with the
293 294 295 296 297
  (spatial) hypotheses `H1 ... Hn` and all intuitionistic hypotheses. The
  spatial hypotheses among `H1 ... Hn` will be consumed, and will not be
  available for subsequent goals. Hypotheses prefixed with a `$` will be framed
  in the goal for the premise. The pattern can be terminated with a `//`, which
  causes `done` to be called to close the goal (after framing).
298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318
- `[-H1 ... Hn]` and `[-H1 ... Hn //]` : the negated forms of the above
  patterns, where the goal for the premise will have all spatial premises except
  `H1 .. Hn`.

- `[> H1 ... Hn]` and `[> H1 ... Hn //]` : like the above patterns, but these
  patterns can only be used if the goal is a modality `M`, in which case
  the goal for the premise will be wrapped in the modality `M`.
- `[> -H1 ... Hn]` and `[> -H1 ... Hn //]` : the negated forms of the above
  patterns.

- `[# $H1 .. $Hn]` and `[# $H1 .. $Hn //]` : generate a goal for a persistent
  premise in which all hypotheses are available. This pattern does not consume
  any hypotheses; all hypotheses are available in the goal for the premise, as
  well in the subsequent goal. The hypotheses `$H1 ... $Hn` will be framed in
  the goal for the premise. These patterns can be terminated with a `//`, which
  causes `done` to be called to close the goal (after framing).
- `[%]` and `[% //]` : generate a Coq goal for a pure premise. This pattern
  does not consume any hypotheses. The pattern can be terminated with a `//`,
  which causes `done` to be called to close the goal.

- `[$]` : solve the premise by framing. It will first repeatedly frame the
319
  spatial hypotheses, and then repeatedly frame the intuitionistic hypotheses.
320 321 322 323 324 325
  Spatial hypothesis that are not framed are carried over to the subsequent
  goal.
- `[> $]` : like the above pattern, but this pattern can only be used if the
  goal is a modality `M`, in which case the goal for the premise will be wrapped
  in the modality `M` before framing.
- `[# $]` : solve the persistent premise by framing. It will first repeatedly
326
  frame the spatial hypotheses, and then repeatedly frame the intuitionistic
327
  hypotheses. This pattern does not consume any hypotheses.
328

329
For example, given:
330

331 332 333
        H : □ P -∗ P 2 -∗ R -∗ x = 0 -∗ Q1 ∗ Q2

One can write:
334

335
        iDestruct ("H" with "[#] [H1 $H2] [$] [% //]") as "[H4 H5]".
336 337


338 339 340 341 342 343
Proof mode terms
================

Many of the proof mode tactics (such as `iDestruct`, `iApply`, `iRewrite`) can
take both hypotheses and lemmas, and allow one to instantiate universal
quantifiers and implications/wands of these hypotheses/lemmas on the fly.
344

345
The syntax for the arguments of these tactics, called _proof mode terms_, is:
346

347
        (H $! t1 ... tn with "spat1 .. spatn")
348

349
Here, `H` can be either a hypothesis or a Coq lemma whose conclusion is
350 351 352
of the shape `P ⊢ Q`. In the above, `t1 ... tn` are arbitrary Coq terms used
for instantiation of universal quantifiers, and `spat1 .. spatn` are
specialization patterns to eliminate implications and wands.
353

354
Proof mode terms can be written down using the following shorthand syntaxes, too:
355

356 357 358
        (H with "spat1 .. spatn")
        (H $! t1 ... tn)
        H
359

360 361 362 363 364
HeapLang tactics
================

If you came here looking for the `wp_` tactics, those are described in the
[HeapLang documentation](HeapLang.md).