Newer
Older
In this changelog, we document "large-ish" changes to Iris that affect even the
way the logic is used on paper. We also document changes in the Coq
development; every API-breaking change should be listed, but not every new
lemma.
**Changes in `algebra`:**
* Add (basic) support for `gset` and `gset_disj` cameras to `set_solver`.
* Rename `sig_{equiv,dist}_alt` into `sig_{equiv,dist}_def` and state these
lemmas using `=` instead of `<->`.
**Changes in `bi`:**
* Add a construction `bi_tc` to create transitive closures of
PROP-level binary relations.
* Use `binder` in notations for big ops. This means one can write things such
as `[∗ map] '(k,_) ↦ '(_,y) ∈ m, ⌜ k = y ⌝`.
**Changes in `proofmode`:**
* The proof mode introduction patterns "<-" and "->" are considered
intuitionistic. This means that tactics such as `iDestruct ... as "->"` will
not dispose of hypotheses to perform the rewrite.
* Remove tactic `iSolveTC` in favor of `tc_solve` in std++.
**LaTeX changes:**
- Rename `\Alloc` to `\AllocN` and `\Ref` to `\Alloc` for better consistency
with the Coq names and to avoid clash with hyperref package.
The following `sed` script helps adjust your code to the renaming (on macOS,
replace `sed` by `gsed`, installed via e.g. `brew install gnu-sed`).
Note that the script is not idempotent, do not run it twice.
```
sed -i -E -f- $(find theories -name "*.v") <<EOF
# iSolveTC
s/iSolveTC\b/tc_solve/g
# _alt -> _def
s/\bsig_equiv_alt\b/sig_equiv_def/g
s/\bsig_dist_alt\b/sig_dist_def/g
The following sed script helps adjust LaTeX documents to these changes:
Note that the script is not idempotent, do not run it twice.
```
sed -i -E -f- *.tex <<EOF
# Alloc & Ref
s/\\Alloc\b/\\AllocN/g
s/\\Ref\b/\\Alloc/g
EOF
```
The highlight of Iris 4.0 is the *later credits* mechanism, which provides a new
way to eliminate later modalities.
This new mechanism complements the existing techniques of taking program steps,
exploiting timelessness, and various modality commuting rules. At each program
step, one obtains a credit `£ 1`, which is an ownable Iris resource. These
credits don't have to be used at the present step, but can be saved up, and used
to eliminate laters at any point in the verification using the fancy update
modality. Later credits are particularly useful in proofs where there is not a
one-to-one correspondence between program steps and later eliminations, for
example, logical atomicity proofs. As a consequence, we have been able to
simplify the definition of logical atomicity by removing the 'laterable'
mechanism.
The later credit mechanism is described in detail in the
[ICFP'22 paper](https://plv.mpi-sws.org/later-credits/) and there is a
[small tutorial](https://gitlab.mpi-sws.org/iris/iris/-/blob/iris-4.0.0/tests/later_credits_paper.v)
in the Iris repository. The
[examples](https://gitlab.mpi-sws.org/iris/examples/) repository contains some
logically atomic case studies that make use of later credits: the counter with a
backup (Section 4 of the later credits paper), as well as the elimination stack,
conditional increment, and RDCSS.
Iris 4.0 supports Coq 8.13 - 8.16.
This release was managed by Ralf Jung, Robbert Krebbers, and Lennard Gäher, with
contributions from Glen Mével, Gregory Malecha, Ike Mulder, Irene Yoon,
Jan-Oliver Kaiser, Jonas Kastberg Hinrichsen, Lennard Gäher, Michael Sammler,
Niklas Mück, Paolo G. Giarrusso, Ralf Jung, Robbert Krebbers, Simon Spies,
and Tej Chajed. Thanks a lot to everyone involved!
**General changes:**
- Rename "unsealing" lemmas from `_eq` to `_unseal`. This particularly
affects `envs_entails_eq`, which is commonly used in the definition of
custom proof mode tactics. All other unsealing lemmas should be internal, so
in principle you should not rely on them.
- Rename `coq-iris-staging` package to `coq-iris-unstable`, and also change the
import path from `iris.staging` to `iris.unstable`.
**Changes in `algebra`:**
* Add some missing algebra functors: `dfrac_agreeRF`, `excl_authURF`, `excl_authRF`,
`frac_authURF`, `frac_authRF`, `ufrac_authURF`, `ufrac_authRF`, `max_prefix_listURF`,
`max_prefix_listRF`, `mono_listURF`, and `mono_listRF`.
* Make validy lemmas for `excl_auth` more consistent with `auth`.
- Rename `excl_auth_frag_validN_op_1_l` into `excl_auth_frag_op_validN` and
`excl_auth_frag_valid_op_1_l` into `excl_auth_frag_op_valid` (similar to
`auth_auth_op_valid`), and make them bi-implications.
- Add `excl_auth_auth_op_validN` and `excl_auth_auth_op_valid`.
* Make validy lemmas for `(u)frac_auth` more consistent with `auth`.
- Remove unidirectional lemmas with `1` fraction `frac_auth_frag_validN_op_1_l`
and `frac_auth_frag_valid_op_1_l`
- Add `frac_auth_frag_op_validN` and `frac_auth_frag_op_valid`, which are
bi-implications with arbitrary fractions.
- Add `ufrac_auth_frag_op_validN` and `ufrac_auth_frag_op_valid`.
Ralf Jung
committed
* Remove `mono_list_lb_is_op` instance for `IsOp' (◯ML l) (◯ML l) (◯ML l)`; we
don't usually have such instances for duplicable resources and it was added by
accident.
**Changes in `bi`:**
* Generalize `big_op` lemmas that were previously assuming `Absorbing`ness of
some assertion: they now take any of (`TCOr`) an `Affine` instance or an
`Absorbing` instance. This breaks uses where an `Absorbing` instance was
provided without relying on TC search (e.g. in `by apply ...`; a possible fix
is `by apply: ...`). (by Glen Mével, Bedrock Systems)
* Change statement of `affinely_True_emp` to also remove the affinely modality.
* Rename `absorbingly_True_emp` to `absorbingly_emp_True` and make statement
consistent with `affinely_True_emp`: `<absorb> emp ⊣⊢ True`.
* Change the notation for atomic updates and atomic accessors (`AU`, `AACC`) to
swap the quantifiers: the first quantifier is logically an existential, the
second a universal, so let's use the appropriate notation. Also use double
quantifiers (`∀∀`, `∃∃`) to make it clear that these are not normal
quantifiers (the latter change was also applied to logically atomic triples).
* Add some lemmas to show properties of functions defined via monotonoe fixpoints:
`least_fixpoint_affine`, `least_fixpoint_absorbing`,
`least_fixpoint_persistent_affine`, `least_fixpoint_persistent_absorbing`,
`greatest_fixpoint_absorbing`.
* Remove `make_laterable` from atomic updates. This relies on Iris now having
support for later credits (see below).
* Add `Fractional` and `AsFractional` instances for `embed` such that the
embedding of something fractional is also fractional. (by Simon Friis Vindum).
**Changes in `proofmode`:**
* Change `iAssumption` to no longer instantiate evar premises with `False`. This
used to occur when the conclusion contains variables that are not in scope of
the evar, thus blocking the default behavior of instantiating the premise with
the conclusion. The old behavior can be emulated with`iExFalso. iExact "H".`
* In `iInduction`, support induction schemes that involve `Forall` and
* Change `iRevert` of a pure hypothesis to generate a magic wand instead of an
implication.
* Change `of_envs` such that when the persistent context is empty, the
persistence modality no longer appears at all. This is a step towards using
the proofmode in logics without a persistence modality.
The lemma `of_envs_alt` shows equivalence with the old version.
* Adjust `IntoWand` instances for non-affine BIs: in many cases where
`iSpecialize`/`iApply` of an implication previously failed, it will now
instead add an `<affine>` modality to the newly generated goal. In some rare
cases it might stop working or add an `<affine>` modality where previously
none was added.
**Changes in `base_logic`:**
* Make the `inG` instances for `libG` fields local, so they are only used inside
the library that defines the `libG`.
* Add infrastructure for supporting later credits, by adding a resource `£ n`
describing ownership of `n` credits that can be eliminated at fancy updates.
+ To retain backwards compatibility with the interaction laws of fancy updates
with the plainly modality (`BiFUpdPlainly`), which are incompatible with
later credits, the logic has a new parameter of type `has_lc`, which is
either `HasLc` or `HasNoLc`. The parameter is an index of the `invGS_gen`
typeclass; the old `invGS` is an alias for `invGS_gen HasLc` so that
developments default to having later credits available. Libraries that want
to be generic over whether credits are available or not, and proofs that
need `BiFUpdPlainly`, need to be changed to use `invGS_gen` rather than
`invGS`.
+ The core soundness lemma `step_fupdN_soundness_gen` similarly takes a `has_lc`
parameter to control how the logic is supposed to be instantiated. The lemma
always generates credits, but they cannot be used in any meaningful way unless
`HasLc` is picked.
* Add discardable fractions `dfrac` to `saved_anything_own`, `saved_prop_own`,
and `saved_pred_own`, so they can be updated. The previous persistent versions
can be recovered with the fraction `DfracDiscarded`. Allocation lemmas now take
a `dq` parameter to define the initial fraction.
* Remove an unused fraction argument to `dfrac_valid_discarded`.
* The definition of the weakest precondition has been changed to generate later credits
(see `base_logic`) for each step:
+ The member `num_laters_per_step` of the `irisGS` class now also determines the number
of later credits that are generated: `S (num_laters_per_step ns)` if `ns` steps
have been taken.
+ The weakest precondition offers credits after a `prim_step` has been proven.
+ All lifting lemmas have been altered to provide credits.
`wp_lift_step_fupdN` provides `S (num_laters_per_step ns)` credits, while all other
lemmas always provide one credit.
* In line with the support for later credits (see `base_logic`), `irisGS_gen`
now also has a `has_lc` parameter and the adequacy statements have been
changed to account for that:
+ The lemma `twp_total` (total adequacy) provides `irisGS_gen HasNoLc`. Clients
of the adequacy proof will need to make sure to be either generic over the
choice of `has_lc` or explicitly opt-out of later credits.
+ The adequacy lemmas for the partial WP, in particular `wp_adequacy`,
`wp_strong_adequacy` and `wp_invariance`, are now available in two flavors:
the old names generate `irisGS` (a short-hand for `irisGS_gen HasLc`); new
lemmas with a `_gen` suffix leave the choice of `has_lc` to the user.
+ The parameter for the stuckness bit `s` in `wp_strong_adequacy{_lc, _no_lc}` has
moved up and is now universally quantified in the lemma instead of being existentially
quantified at the Iris-level. For clients that already previously quantified over `s`
at the Coq level, the only required change should be to remove the instantiation
of the existential quantifier.
**Changes in `iris_heap_lang`:**
* Change the `num_laters_per_step` of `heap_lang` to `λ n, n`, signifying that
each step of the weakest precondition strips `n` laters, where `n` is the
number of steps taken so far. This number is tied to ghost state in the state
interpretation, which is exposed, updated, and used with new lemmas
`wp_lb_init`, `wp_lb_update`, and `wp_step_fupdN_lb`. (by Jonas Kastberg Hinrichsen)
* Make pattern argument of `wp_pure` tactic optional (defaults to wildcard
pattern, matching all redexes).
* In line with the support for later credits (see `base_logic`), the tactic
`wp_pure` now takes an optional parameter `credit:"H"` which generates a
hypothesis `H` for a single later credit `£ 1` that can be eliminated using
`lc_fupd_elim_later`.
The typeclass `heapGS_gen` now takes an additional `has_lc` parameter, and
`heapGS` is a short-hand for `heapGS_gen HasLc`. The adequacy statements for
HeapLang have been changed accordingly:
+ `heap_adequacy` provides `heapGS`, thus enabling the use of later credits.
This precludes usage of the laws in `BiFUpdPlainly` in the HeapLang instance of Iris.
The following `sed` script helps adjust your code to the renaming (on macOS,
replace `sed` by `gsed`, installed via e.g. `brew install gnu-sed`).
Note that the script is not idempotent, do not run it twice.
```
sed -i -E -f- $(find theories -name "*.v") <<EOF
# excl_auth
s/\bexcl_auth_frag_validN_op_1_l\b/excl_auth_frag_op_validN/g
s/\bexcl_auth_frag_valid_op_1_l\b/excl_auth_frag_op_valid/g
# staging → unstable
s/\biris\.staging\b/iris.unstable/g
# plus → add
s/\blaterN_plus\b/laterN_add/g
s/\bpos_op_plus\b/pos_op_add/g
## Iris 3.6.0 (2022-01-22)
The highlights and most notable changes of this release are:
* Coq 8.15 is now supported, while Coq 8.13 and Coq 8.14 remain supported.
Coq 8.12 is no longer supported.
* Support for discardable fractions (`dfrac`) has been added to `gmap_view`
authoritative elements, and to the `mono_nat` library. See below for other
`dfrac`-related changes.
* A new `mono_list` algebra provides monotonically growing lists with an
exclusive authoritative element and persistent prefix witnesses. See
`iris/algebra/lib/mono_list.v` for details. An experimental logic-level
library wrapping the algebra is available at
`iris_staging/base_logic/mono_list.v`; if you use it, please give feedback on
the tracking issue
[iris/iris#439](https://gitlab.mpi-sws.org/iris/iris/-/issues/439).
This release was managed by Ralf Jung, Robbert Krebbers, and Tej Chajed, with
contributions from Dan Frumin, Jonas Kastberg Hinrichsen, Lennard Gäher,
Matthieu Sozeau, Michael Sammler, Paolo G. Giarrusso, Ralf Jung, Robbert
Krebbers, Simon Friis Vindum, Tej Chajed, and Vincent Siles. Thanks a lot to
everyone involved!
**Changes in `algebra`**
* Define non-expansive instance for `dom`. This, in particular, makes it
possible to `iRewrite` below `dom` (even if the `dom` appears in `⌜ _ ⌝`).
* Generalize the authorative elements of `gmap_view` to be parameterized by a
[discardable fraction](iris/algebra/dfrac.v) (`dfrac`) instead of a fraction
(`frac`). Lemmas affected by this have been renamed such that the "frac" in
their name has been changed into "dfrac". (by Simon Friis Vindum)
* Change `ufrac_auth` notation to not use curly braces, since these fractions do
not behave like regular fractions (and cannot be made `dfrac`).
Old: `●U{q} a`, `◯U{q} b`; new: `●U_q a`, `◯U_q b`.
* Equip `frac_agree` with support for `dfrac` and rename it to `dfrac_agree`.
The old `to_frac_agree` and its lemmas still exist, except that the
`frac_agree_op_valid` lemmas are made bi-directional.
* Rename typeclass instance `Later_inj` -> `Next_inj`.
* Remove `view_auth_frac_op`, `auth_auth_frac_op`, `gmap_view_auth_frac_op`; the
corresponding `dfrac` lemmas can be used instead (together with `dfrac_op_own`
if needed).
* Equip `mono_nat` algebra with support for `dfrac`, make API more consistent,
and add notation for algebra elements. See `iris/algebra/lib/mono_nat.v` for
details. This affects some existing terms and lemmas:
- `mono_nat_auth` now takes a `dfrac`, but the recommendation is to port to the notation.
- `mono_nat_lb_op`: direction of equality is swapped.
- `mono_nat_auth_frac_op`, `mono_nat_auth_frac_op_valid`,
`mono_nat_auth_frac_valid`, `mono_nat_both_frac_valid`: use `dfrac` variant
instead.
* Add `mono_list` algebra for monotonically growing lists with an exclusive
authoritative element and persistent prefix witnesses. See
`iris/algebra/lib/mono_list.v` for details.
* Rename `least_fixpoint_ind` into `least_fixpoint_iter`,
rename `greatest_fixpoint_coind` into `greatest_fixpoint_coiter`,
rename `least_fixpoint_strong_ind` into `least_fixpoint_ind`,
add lemmas `least_fixpoint_{ind_wf, ne', strong_mono}`, and
add lemmas `greatest_fixpoint_{coind, paco, ne', strong_mono}`.
* Move `persistently_forall_2` (`∀ <pers> ⊢ <pers> ∀`) out of the BI interface
into a new typeclass, `BiPersistentlyForall`. The BI interface instead just
demands the equivalent property for conjunction (`(<pers> P) ∧ (<pers> Q) ⊢
<pers> (P ∧ Q)`). This enables the IPM to support logics where the
persistently modality is defined with an existential quantifier. This also
necessitates removing `persistently_impl_plainly` from `BiPlainly` into a new
typeclass `BiPersistentlyImplPlainly`.
Proofs that are generic in `PROP` might have to add those new classes as
assumptions to remain compatible, and code that instantiates the BI interface
* Make `frame_fractional` not an instance any more; instead fractional
propositions that want to support framing are expected to register an
appropriate instance themselves. HeapLang and gen_heap `↦` still support
framing, but the other fractional propositions in Iris do not.
**Changes in `heap_lang`:**
* The `is_closed_expr` predicate is formulated in terms of a
set of binders (as opposed to a list of binders).
The following `sed` script helps adjust your code to the renaming (on macOS,
replace `sed` by `gsed`, installed via e.g. `brew install gnu-sed`).
Note that the script is not idempotent, do not run it twice.
```
sed -i -E -f- $(find theories -name "*.v") <<EOF
# least/greatest fixpoint renames
s/\bleast_fixpoint_ind\b/least_fixpoint_iter/g
s/\bgreatest_fixpoint_coind\b/greatest_fixpoint_coiter/g
s/\bleast_fixpoint_strong_ind\b/least_fixpoint_ind/g
# gmap_view renames from frac to dfrac
s/\bgmap_view_(auth|both)_frac_(op_invN|op_inv|op_inv_L|valid|op_validN|op_valid|op_valid_L)\b/gmap_view_\1_dfrac_\2/g
s/\bgmap_view_persist\b/gmap_view_frag_persist/g
# frac_agree with dfrac
s/\bfrac_agreeR\b/dfrac_agreeR/g
The highlights and most notable changes of this release are:
* Coq 8.14 is now supported, while Coq 8.12 and Coq 8.13 remain supported.
* The proof mode now has native support for pure names `%H` in intro patterns,
without installing
[iris/string-ident](https://gitlab.mpi-sws.org/iris/string-ident). If you had
the plugin installed, to migrate simply uninstall the plugin and stop
importing it.
* The proof mode now supports destructing existentials with the `"[%x ...]"`
pattern.
* `iMod` and `iModIntro` now report an error message for mask mismatches.
* Performance improvements for the proof mode in `iFrame` in non-affine
logics, `iPoseProof`, and `iDestruct` (by Paolo G. Giarrusso, Bedrock Systems,
and Armaël Guéneau).
* The new `ghost_map` logic-level library supports a ghost `gmap K V` with an
authoritative view and per-element points-to facts written `k ↪[γ] w`.
* Weakest preconditions now support a flexible number of laters per
physical step of the operational semantics. See merge request
[!585](https://gitlab.mpi-sws.org/iris/iris/-/merge_requests/595) (by
* HeapLang now has an atomic `Xchg` (exchange) operation (by Simon Hudon,
Google).
This release was managed by Ralf Jung, Robbert Krebbers, and Tej Chajed, with
contributions from Amin Timany, Armaël Guéneau, Dan Frumin, Dmitry Khalanskiy,
Hoang-Hai Dang, Jacques-Henri Jourdan, Lennard Gäher, Michael Sammler, Paolo G.
Giarrusso, Ralf Jung, Robbert Krebbers, Simon Friis Vindum, Simon Hudon, Tej
Chajed, and Yusuke Matsushita. Thanks a lot to everyone involved!
**Changes in `algebra`:**
* Generalize the authorative elements of the `view`, `auth` and `gset_bij`
cameras to be parameterized by a [discardable fraction](iris/algebra/dfrac.v)
(`dfrac`) instead of a fraction (`frac`). Normal fractions are now denoted
`●{#q} a` and `●V{#q} a`. Lemmas affected by this have been renamed such that
the "frac" in their name has been changed into "dfrac". (by Simon Friis Vindum)
* Generalize `namespace_map` to `reservation_map` which enhances `gmap positive
A` with a notion of 'tokens' that enable allocating a particular name in the
map. See [algebra.reservation_map](iris/algebra/reservation_map.v) for further
information.
* Add `dyn_reservation_map` which further extends `reservation_map` with the
ability to dynamically allocate an infinite set of tokens. This is useful to
perform synchronized allocation of the same name in two maps/APIs without
dedicated support from one of the involved maps/APIs. See
[algebra.dyn_reservation_map](iris/algebra/dyn_reservation_map.v) for further
information.
* Demote the Camera structure on `list` to `iris_staging` since its composition
is not very well-behaved.
* Extend `gmap_view` with lemmas for "big" operations on maps.
Jacques-Henri Jourdan
committed
* Typeclasses instances triggering a canonical structure search such as `Equiv`,
`Dist`, `Op`, `Valid`, `ValidN`, `Unit`, `PCore` now use an `Hint Extern`
based on `refine` instead of `apply`, in order to use Coq's newer unification
algorithm.
* Set `Hint Mode` for the classes `OfeDiscrete`, `Dist`, `Unit`, `CmraMorphism`,
* Set `Hint Mode` for the stdpp class `Equiv`. This might require few spurious
type annotations until
[Coq bug #14441](https://github.com/coq/coq/issues/14441) is fixed.
* Add `max_prefix_list` RA on lists whose composition is only defined when one
operand is a prefix of the other. The result is the longer list.
**Changes in `bi`:**
* Add new lemmas `big_sepM2_delete_l` and `big_sepM2_delete_r`.
* Rename `big_sepM2_lookup_1` → `big_sepM2_lookup_l` and
`big_sepM2_lookup_2` → `big_sepM2_lookup_r`.
* Add lemmas for swapping nested big-ops: `big_sep{L,M,S,MS}_sep{L,M,S,MS}`.
Ralf Jung
committed
* Rename `big_sep{L,L2,M,M2,S}_intuitionistically_forall` →
`big_sep{L,L2,M,M2,S}_intro`, and `big_orL_lookup` → `big_orL_intro`.
* Rename `bupd_forall` to `bupd_plain_forall`, and add
`{bupd,fupd}_{and,or,forall,exist}`.
* Decouple `Wp` and `Twp` typeclasses from the `program_logic.language`
interface. The typeclasses are now parameterized over an expression and a
value type, instead of a language. This requires extra type annotations or
explicit coercions in a few cases, in particular `WP v {{ Φ }}` must now be
written `WP (of_val v) {{ Φ }}`.
- Adjust definition such that `Laterable P` iff `P ⊢ make_laterable P`.
As a consequence, `make_laterable_elim` got weaker: elimination now requires
an except-0 modality (`make_laterable P -∗ ◇ P`).
- Add `iModIntro` support for `make_laterable`.
- Use `□`/`-∗` instead of `<pers>`/`→`.
- Strengthen to ensure that functions for recursive calls are non-expansive.
* Add `big_andM` (big conjunction on finite maps) with lemmas similar to `big_andL`.
* Add transitive embedding that constructs an embedding of `PROP1` into `PROP3`
by combining the embeddings of `PROP1` into `PROP2` and `PROP2` into `PROP3`.
This construct is *not* declared as an instance to avoid TC search divergence.
* Improve notation printing around magic wands, view shifts, `WP`, Texan
triples, and logically atomic triples.
* Slight change to the `AACC` notation for atomic accessors (which is usually
only printed, not parsed): added a `,` before `ABORT`, for consistency with `COMM`.
* Add the lemmas `big_sepM_impl_strong` and `big_sepM_impl_dom_subseteq` that
generalize the existing `big_sepM_impl` lemma. (by Simon Friis Vindum)
* Add new instance `fractional_big_sepL2`. (by Paolo G. Giarrusso, BedRock Systems)
* Add support for pure names `%H` in intro patterns. This is now natively
supported whereas the previous experimental support required installing
https://gitlab.mpi-sws.org/iris/string-ident. (by Tej Chajed)
* Add support for destructing existentials with the intro pattern `[%x ...]`.
* `iMod`/`iModIntro` show proper error messages when they fail due to mask
mismatches. To support this, the proofmode typeclass `FromModal` now takes an
additional pure precondition.
* Fix performance of `iFrame` in logics without `BiAffine`.
To adjust your code if you use such logics and define `Frame` instances,
ensure these instances to have priority at least 2: they should have either at
least 2 (non-dependent) premises, or an explicit priority.
References: docs for `frame_here_absorbing` in
[iris/proofmode/frame_instances.v](iris/proofmode/frame_instances.v) and
https://coq.inria.fr/refman/addendum/type-classes.html#coq:cmd.Instance. (by
Paolo G. Giarrusso, BedRock Systems)
* Rename the main entry point module for the proofmode from
`iris.proofmode.tactics` to `iris.proofmode.proofmode`. Under normal
circumstances, this should be the only proofmode file you need to import.
* Improve performance of the `iIntoEmpValid` tactic used by `iPoseProof`,
especially in the case of large goals and lemmas with many forall quantifiers.
* Improve performance of the `iDestruct` tactic, by using user-provided names
more eagerly in order to avoid later calls to `iRename`.
(by Armaël Guéneau)
**Changes in `bi`:**
* Add lemmas characterizing big-ops over pure predicates (`big_sep*_pure*`).
* Move `BiAffine`, `BiPositive`, `BiLöb`, and `BiPureForall` from
`bi.derived_connectives` to `bi.extensions`.
* Strengthen `persistent_fractional` to support propositions that are persistent
and either affine or absorbing. (by Paolo G. Giarrusso, BedRock Systems)
**Changes in `base_logic`:**
* Add `ghost_map`, a logic-level library for a `gmap K V` with an authoritative
view and per-element points-to facts written `k ↪[γ] w`.
* Generalize the soundness lemma of the base logic `step_fupdN_soundness`.
It applies even if invariants stay open accross an arbitrary number of laters.
* Rename those `*G` typeclasses that must be global singletons to `*GS`, and
their corresponding `preG` class to `GpreS`. Affects `invG`, `irisG`,
`gen_heapG`, `inv_heapG`, `proph_mapG`, `ownPG`, `heapG`.
Jacques-Henri Jourdan
committed
**Changes in `program_logic`:**
* Change definition of weakest precondition to use a variable number of laters
(i.e., logical steps) for each physical step of the operational semantics,
depending on the number of physical steps executed since the begining of the
execution of the program. See merge request [!595](https://gitlab.mpi-sws.org/iris/iris/-/merge_requests/595).
This implies several API-breaking changes, which can be easily fixed in client
formalizations in a backward compatible manner as follows:
Jacques-Henri Jourdan
committed
- Ignore the new parameter `ns` in the state interpretation, which
corresponds to a step counter.
- Use the constant function "0" for the new field `num_laters_per_step` of
`irisG`.
- Use `fupd_intro _ _` for the new field `state_interp_mono` of `irisG`.
- Some proofs using lifting lemmas and adequacy theorems need to be adapted
to ignore the new step counter.
* Remove `wp_frame_wand_l`; add `wp_frame_wand` as more symmetric replacement.
* Swap the polarity of the mask in logically atomic triples, so that it matches
regular `WP` masks.
Jacques-Henri Jourdan
committed
**Changes in `heap_lang`:**
* Rename `Build_loc` constructor for `loc` type to `Loc`.
* Add atomic `Xchg` ("exchange"/"swap") operation. (by Simon Hudon, Google LLC)
The following `sed` script helps adjust your code to the renaming (on macOS,
replace `sed` by `gsed`, installed via e.g. `brew install gnu-sed`).
Note that the script is not idempotent, do not run it twice.
```
sed -i -E -f- $(find theories -name "*.v") <<EOF
s/\b(auth|view)_(auth|both|update)_frac_(is_op|op_invN|op_inv|inv_L|validN|op_validN|valid|op_valid|valid_2|valid_discrete|includedN|included|alloc|validI|validI_2|validI_1|validI|)\b/\1_\2_dfrac_\3/g
s/\bgset_bij_auth_frac_(\w*)\b/gset_bij_auth_dfrac_\1/g
s/\bgset_bij_auth_empty_frac_valid\b/gset_bij_auth_empty_dfrac_valid/g
s/\bbij_both_frac_valid\b/bij_both_dfrac_valid/g
s/\bbig_sepM2_lookup_1\b/big_sepM2_lookup_l/g
s/\bbig_sepM2_lookup_2\b/big_sepM2_lookup_r/g
Ralf Jung
committed
# big_*_intro
s/\bbig_sep(L|L2|M|M2|S)_intuitionistically_forall\b/big_sep\1_intro/g
s/\bbig_orL_lookup\b/big_orL_intro/g
s/\bbupd_forall\b/bupd_plain_forall/g
# "global singleton" rename
s/\b(inv|iris|(gen|inv)_heap|(Gen|Inv)Heap|proph_map|ProphMap|[oO]wnP|[hH]eap)G\b/\1GS/g
s/\b([iI]nv|iris|(gen|inv)_heap|(Gen|Inv)Heap|proph_map|ProphMap|[oO]wnP|[hH]eap)PreG\b/\1GpreS/g
# iris.proofmode.tactics → iris.proofmode.proofmode
s/\bproofmode\.tactics\b/proofmode.proofmode/
s/(From +iris\.proofmode +Require +(Import|Export).*)\btactics\b/\1proofmode/
# iris_invG → iris_invGS
s/\biris_invG\b/iris_invGS/g
The highlights and most notable changes of this release are as follows:
* Coq 8.13 is now supported; the old Coq 8.9 and Coq 8.10 are not supported any
more.
* The new `view` RA construction generalizes `auth` to user-defined abstraction
relations. (thanks to Gregory Malecha for the inspiration)
* The new `dfrac` RA extends `frac` (fractions `0 < q ≤ 1`) with support for
"discarding" some part of the fraction in exchange for a persistent witness
that discarding has happened. This can be used to easily generalize fractional
permissions with support for persistently owning "any part" of the resource.
(by Simon Friis Vindum)
* The new `gmap_view` RA provides convenient lemmas for ghost ownership
of heap-like structures with an "authoritative" view. Thanks to `dfrac`, it
supports both exclusive (mutable) and persistent (immutable) ownership of
individual map elements.
* With this release we are beginning to provide logic-level abstractions for
ghost state, which have the advantage that the user does not have to directly
interact with RAs to use them.
- `ghost_var` provides a logic-level abstraction of ghost variables: a mutable
"variable" with fractional ownership.
- `mono_nat` provides a "monotone counter" with a persistent witnesses
representing a lower bound of the current counter value. (by Tej Chajed)
- `gset_bij` provides a monotonically growing partial bijection; this is
useful in particular when building binary logical relations for languages
with a heap.
* HeapLang provides a persistent read-only points-to assertion `l ↦□ v`.
(by Simon Friis Vindum)
* We split Iris into multiple opam packages: `coq-iris` no longer contains
HeapLang, which is now in a separate package `coq-iris-heap-lang`. The two
packages `coq-iris-deprecated` (for old modules that we eventually plan to
remove entirely) and `coq-iris-staging` (for new modules that are not yet
ready for prime time) exist only as development versions, so they are not part
of this release.
* The proofmode now does a better job at picking reasonable names when moving
variables into the Coq context without a name being explicitly given by the
user. However, the exact variable names remain unspecified. (by Tej Chajed)
Further details are given in the changelog below.
This release of Iris was managed by Ralf Jung and Robbert Krebbers, with
contributions by Arthur Azevedo de Amorim, Dan Frumin, Enrico Tassi, Hai Dang,
Michael Sammler, Paolo G. Giarrusso, Rodolphe Lepigre, Simon Friis Vindum, Tej
Chajed, and Yusuke Matsushita. Thanks a lot to everyone involved!
**Changes in `algebra`:**
* Add constructions to define a camera through restriction of the validity predicate
(`iso_cmra_mixin_restrict`) and through an isomorphism (`iso_cmra_mixin`).
* Add a `frac_agree` library which encapsulates `frac * agree A` for some OFE
`A`, and provides some useful lemmas.
* Add the view camera `view`, which generalizes the authoritative camera
`auth` by being parameterized by a relation that relates the authoritative
element with the fragments.
* Add the camera of discardable fractions `dfrac`. This is a generalization of
the normal fractional camera.
See [algebra.dfrac](iris/algebra/dfrac.v) for further information.
* Add `gmap_view`, a camera providing a "view of a `gmap`". The authoritative
element is any `gmap`; the fragment provides fractional ownership of a single
key, including support for persistent read-only ownership through `dfrac`.
See [algebra.lib.gmap_view](iris/algebra/lib/gmap_view.v) for further information.
* Add `mono_nat`, a wrapper for `auth max_nat`. The result is an authoritative
`nat` where a fragment is a lower bound whose ownership is persistent.
See [algebra.lib.mono_nat](iris/algebra/lib/mono_nat.v) for further information.
* Add the `gset_bij` resource algebra for monotone partial bijections.
See [algebra.lib.gset_bij](iris/algebra/lib/gset_bij.v) for further information.
* Rename `agree_op_inv'` → `to_agree_op_inv`,
`agree_op_invL'` → `to_agree_op_inv_L`, and add `to_agree_op_invN`.
* Rename `auth_auth_frac_op_invL` → `auth_auth_frac_op_inv_L`,
`excl_auth_agreeL` → `excl_auth_agree_L`,
`frac_auth_agreeL` → `frac_auth_agree_L`, and
`ufrac_auth_agreeL` → `ufrac_auth_agree_L`.
Robbert Krebbers
committed
* Fix direction of `auth_auth_validN` to make it consistent with similar lemmas,
e.g., `auth_auth_valid`. The direction is now `✓{n} (● a) ↔ ✓{n} a`.
* Rename `auth_both_valid` to `auth_both_valid_discrete` and
`auth_both_frac_valid` to `auth_both_frac_valid_discrete`. The old name is
used for new, stronger lemmas that do not assume discreteness.
* Redefine the authoritative camera in terms of the view camera. As part of this
change, we have removed lemmas that leaked implementation details. Hence, the
only way to construct elements of `auth` is via the elements `●{q} a` and
`◯ b`. The constructor `Auth`, and the projections `auth_auth_proj` and
`auth_frag_proj` no longer exist. Lemmas that referred to these constructors
have been removed, in particular: `auth_equivI`, `auth_validI`,
`auth_included`, `auth_valid_discrete`, and `auth_both_op`. For validity, use
`auth_auth_valid*`, `auth_frag_valid*`, or `auth_both_valid*` instead.
* Rename `auth_update_core_id` into `auth_update_frac_alloc`.
* Rename `cmra_monotone_valid` into `cmra_morphism_valid` (this rename was
forgotten in !56).
* Move the `*_validI` and `*_equivI` lemmas to a new module, `base_logic.algebra`.
That module is exported by `base_logic.base_logic` so it should usually be
available everywhere without further changes.
* The authoritative fragment `✓ (◯ b : auth A)` is no longer definitionally
equal to `✓ b`.
* Change `*_valid` lemma statements involving fractions to use `Qp` addition and
Ralf Jung
committed
inequality instead of RA composition and validity (also in `base_logic` and
the higher layers).
* Strengthen `cmra_op_discrete` to assume only `✓{0} (x1 ⋅ x2)` instead of `✓
(x1 ⋅ x2)`.
* Rename the types `ofeT`→`ofe`, `cmraT`→`cmra`, `ucmraT`→`ucmra`, and the
constructors `OfeT`→`Ofe`, `CmraT`→`Cmra`, and `UcmraT`→`Ucmra` since the `T`
suffix is not needed. This change makes these names consistent with `bi`,
which also does not have a `T` suffix.
* Rename typeclass instances of CMRA operational typeclasses (`Op`, `Core`,
`PCore`, `Valid`, `ValidN`, `Unit`) to have a `_instance` suffix, so that
their original names are available to use as lemma names.
* Rename `frac_valid'`→`frac_valid`, `frac_op'`→`frac_op`,
`ufrac_op'`→`ufrac_op`, `coPset_op_union` → `coPset_op`, `coPset_core_self` →
`coPset_core`, `gset_op_union` → `gset_op`, `gset_core_self` → `gset_core`,
`gmultiset_op_disj_union` → `gmultiset_op`, `gmultiset_core_empty` →
`gmultiset_core`, `nat_op_plus` → `nat_op`, `max_nat_op_max` →
`max_nat_op`. Those names were previously blocked by typeclass instances.
**Changes in `bi`:**
* Add big op lemmas `big_op{L,L2,M,M2,S}_intuitionistically_forall` and
`big_sepL2_forall`, `big_sepMS_forall`, `big_sepMS_impl`, and `big_sepMS_dup`.
* Add lemmas to big-ops that provide ownership of a single element and permit
changing the quantified-over predicate when re-assembling the big-op:
`big_sepL_lookup_acc_impl`, `big_sepL2_lookup_acc_impl`,
`big_sepM_lookup_acc_impl`, `big_sepM2_lookup_acc_impl`,
`big_sepS_elem_of_acc_impl`, `big_sepMS_elem_of_acc_impl`.
* Add lemmas `big_sepM_filter'` and `big_sepM_filter` matching the corresponding
`big_sepS` lemmas.
* Add lemmas for big-ops of magic wands: `big_sepL_wand`, `big_sepL2_wand`,
`big_sepM_wand`, `big_sepM2_wand`, `big_sepS_wand`, `big_sepMS_wand`.
* Add notation `¬ P` for `P → False` to `bi_scope`.
* Add `fupd_mask_intro` which can be conveniently `iApply`ed to goals of the
form `|={E1,E2}=>` to get rid of the `fupd` in the goal if `E2 ⊆ E1`. The
lemma `fupd_mask_weaken Enew` can be `iApply`ed to shrink the first mask to
`Enew` without getting rid of the modality; the same effect can also be
obtained slightly more conveniently by using `iMod` with `fupd_mask_subseteq
Enew`. To make the new names work, rename some existing lemmas:
`fupd_intro_mask` → `fupd_mask_intro_subseteq`,
`fupd_intro_mask'` → `fupd_mask_subseteq` (implicit arguments also changed
here), `fupd_mask_weaken` → `fupd_mask_intro_discard`. Remove `fupd_mask_same`
since it was unused and obscure. In the `BiFUpd` axiomatization, rename
`bi_fupd_mixin_fupd_intro_mask` to `bi_fupd_mixin_fupd_mask_subseteq` and
weaken the lemma to be specifically about `emp` (the stronger version can be
derived).
* Remove `bi.tactics` with tactics that predate the proofmode (and that have not
been working properly for quite some time).
* Strengthen `persistent_sep_dup` to support propositions that are persistent
and either affine or absorbing.
* Fix the statement of the lemma `fupd_plainly_laterN`; the old lemma was a
duplicate of `fupd_plain_laterN`.
* Strengthen `big_sepL2_app_inv` by weakening a premise (it is sufficient for
one of the two pairs of lists to have equal length).
* Rename `equiv_entails` → `equiv_entails_1_1`,
`equiv_entails_sym` → `equiv_entails_1_2`, and `equiv_spec` → `equiv_entails`.
* Remove the laws `pure_forall_2 : (∀ a, ⌜ φ a ⌝) ⊢ ⌜ ∀ a, φ a ⌝` from the BI
interface and factor it into a type class `BiPureForall`.
**Changes in `proofmode`:**
* The proofmode now preserves user-supplied names for existentials when using
`iDestruct ... as (?) "..."`. This is backwards-incompatible if you were
relying on the previous automatic names (which were always "H", possibly
freshened). It also requires some changes if you were implementing `IntoExist`
yourself, since the typeclass now forwards names. If your instance transforms
one `IntoExist` into another, you can generally just forward the name from the
premise.
* The proofmode also preserves user-supplied names in `iIntros`, for example
with `iIntros (?)` and `iIntros "%"`, as described for destructing
existentials above. As part of this change, it now uses a base name of `H` for
pure facts rather than the previous default of `a`. This also requires some
changes if you were implementing `FromForall`, in order to forward names.
* Make `iFrame` "less" smart w.r.t. clean up of modalities. It now consistently
removes the modalities `<affine>`, `<absorbing>`, `<persistent>` and `□` only
if the result after framing is `True` or `emp`. In particular, it no longer
removes `<affine>` if the result after framing is affine, and it no longer
removes `□` if the result after framing is intuitionistic.
* Allow framing below an `<affine>` modality if the hypothesis that is framed is
affine. (Previously, framing below `<affine>` was only possible if the
hypothesis that is framed resides in the intuitionistic context.)
* Add Coq side-condition `φ` to class `ElimAcc` (similar to what we already had
for `ElimInv` and `ElimModal`).
* Add a tactic `iSelect pat tac` (similar to `select` in std++) which runs the
tactic `tac H` with the name `H` of the last hypothesis of the intuitionistic
or spatial context matching `pat`. The tactic `iSelect` is used to implement:
+ `iRename select (pat)%I into name` which renames the matching hypothesis,
+ `iDestruct select (pat)%I as ...` which destructs the matching hypothesis,
+ `iClear select (pat)%I` which clears the matching hypothesis,
+ `iRevert select (pat)%I` which reverts the matching hypothesis,
+ `iFrame select (pat)%I` which cancels the matching hypothesis.
**Changes in `base_logic`:**
* Add a `ghost_var` library that provides (fractional) ownership of a ghost
variable of arbitrary `Type`.
* Define a ghost state library on top of the `mono_nat` resource algebra.
See [base_logic.lib.mono_nat](iris/base_logic/lib/mono_nat.v) for further
information.
* Define a ghost state library on top of the `gset_bij` resource algebra.
See [base_logic.lib.gset_bij](iris/base_logic/lib/gset_bij.v) for further
information.
* Extend the `gen_heap` library with read-only points-to assertions using
[discardable fractions](iris/algebra/dfrac.v).
+ The `mapsto` connective now takes a `dfrac` rather than a `frac` (i.e.,
positive rational number `Qp`).
+ The notation `l ↦{ dq } v` is generalized to discardable fractions
`dq : dfrac`.
+ The new notation `l ↦{# q} v` is used for a concrete fraction `q : frac`
(e.g., to enable writing `l ↦{# 1/2} v`).
+ The new notation `l ↦□ v` is used for the discarded fraction. This
persistent proposition provides read-only access to `l`.
+ The lemma `mapsto_persist : l ↦{dq} v ==∗ l ↦□ v` is used for making the
location `l` read-only.
+ See the [changes to HeapLang](https://gitlab.mpi-sws.org/iris/iris/-/merge_requests/554)
for an indication on how to adapt your language.
+ See the [changes to iris-examples](https://gitlab.mpi-sws.org/iris/examples/-/commit/a8425b708ec51d918d5cf6eb3ab6fde88f4e2c2a)
for an indication on how to adapt your development. In particular, instead
of `∃ q, l ↦{q} v` you likely want to use `l ↦□ v`, which has the advantage
of being persistent (rather than just duplicable).
* Change type of some ghost state lemmas (mostly about allocation) to use `∗`
instead of `∧` (consistent with our usual style). This affects the following
lemmas: `own_alloc_strong_dep`, `own_alloc_cofinite_dep`, `own_alloc_strong`,
`own_alloc_cofinite`, `own_updateP`, `saved_anything_alloc_strong`,
`saved_anything_alloc_cofinite`, `saved_prop_alloc_strong`,
`saved_prop_alloc_cofinite`, `saved_pred_alloc_strong`,
`saved_pred_alloc_cofinite`, `auth_alloc_strong`, `auth_alloc_cofinite`,
`auth_alloc`.
* Change `uPred_mono` to only require inclusion at the smaller step-index.
* Put `iProp`/`iPreProp`-isomorphism into the `own` construction. This affects
clients that define higher-order ghost state constructions. Concretely, when
defining an `inG`, the functor no longer needs to be applied to `iPreProp`,
but should be applied to `iProp`. This avoids clients from having to push
through the `iProp`/`iPreProp`-isomorphism themselves, which is now handled
once and for all by the `own` construction.
* Rename `gen_heap_ctx` to `gen_heap_interp`, since it is meant to be used in
the state interpretation of WP and since `_ctx` is elsewhere used as a suffix
indicating "this is a persistent assumption that clients should always have in
their context". Likewise, rename `proph_map_ctx` to `proph_map_interp`.
* Move `uPred.prod_validI`, `uPred.option_validI`, and
`uPred.discrete_fun_validI` to the new `base_logic.algebra` module. That
module is exported by `base_logic.base_logic` so these names are now usually
available everywhere, and no longer inside the `uPred` module.
* Remove the `gen_heap` notations `l ↦ -` and `l ↦{q} -`. They were barely used
and looked very confusing in context: `l ↦ - ∗ P` looks like a magic wand.
* Change `gen_inv_heap` notation `l ↦□ I` to `l ↦_I □`, so that `↦□` can be used
by `gen_heap`.
* Strengthen `mapsto_valid_2` conclusion from `✓ (q1 + q2)%Qp` to
`⌜✓ (q1 + q2)%Qp ∧ v1 = v2⌝`.
* Change `gen_heap_init` to also return ownership of the points-to facts for the
initial heap.
Ralf Jung
committed
* Rename `mapsto_mapsto_ne` to `mapsto_frac_ne`, and add a simpler
`mapsto_ne` that does not require reasoning about fractions.
* Deprecate the `auth` and `sts` modules. These were logic-level wrappers around
the underlying RAs; as far as we know, they are unused since they were not
flexible enough for practical use.
* Deprecate the `viewshift` module, which defined a binary view-shift connective
with an implicit persistence modality. It was unused and too easily confused
with `={_}=∗`, the binary view-shift (fancy update) *without* a persistence
modality.
**Changes in `program_logic`:**
* `wp_strong_adequacy` now applies to an initial state with multiple
threads instead of only a single thread. The derived adequacy lemmas
are unchanged.
* `pure_exec_fill` is no longer registered as an instance for `PureExec`, to
avoid TC search attempting to apply this instance all the time.
* Merge `wp_value_inv`/`wp_value_inv'` into `wp_value_fupd`/`wp_value_fupd'` by
* Generalize HeapLang's `mapsto` (`↦`), `array` (`↦∗`), and atomic heap
connectives to discardable fractions. See the CHANGELOG entry in the category
`base_logic` for more information.
* Opening an invariant or eliminating a mask-changing update modality around a
non-atomic weakest precondition creates a side-condition `Atomic ...`.
Before, this would fail with the unspecific error "iMod: cannot eliminate
* In `Ectx_step` and `step_atomic`, mark the parameters that are determined by
the goal as implicit.
* Deprecate the `hoare` module to prevent accidental usage; the recommended way
to write Hoare-style specifications is to use Texan triples.
**Changes in `heap_lang`:**
* `wp_pures` now turns goals of the form `WP v {{ Φ }}` into `Φ v`.
* Fix `wp_bind` in case of a NOP (i.e., when the given expression pattern is
already at the top level).
* The `wp_` tactics now preserve the possibility of doing a fancy update when
the expression reduces to a value.
* Move `IntoVal`, `AsVal`, `Atomic`, `AsRecV`, and `PureExec` instances to their
own file [heap_lang.class_instances](iris_heap_lang/class_instances.v).
* Move `inv_head_step` tactic and `head_step` auto hints (now part of new hint
database `head_step`) to [heap_lang.tactics](iris_heap_lang/tactics.v).
* The tactic `wp_apply` no longer performs `wp_pures` before applying the given
lemma. The new tactic `wp_smart_apply` repeatedly performs single `wp_pure`
steps until the lemma matches the goal.
The following `sed` script helps adjust your code to the renaming (on macOS,
replace `sed` by `gsed`, installed via e.g. `brew install gnu-sed`).
Note that the script is not idempotent, do not run it twice.
```
sed -i -E -f- $(find theories -name "*.v") <<EOF
# agree and L suffix renames
s/\bagree_op_inv'/to_agree_op_inv/g
s/\bagree_op_invL'/to_agree_op_inv_L/g
s/\bauth_auth_frac_op_invL\b/auth_auth_frac_op_inv_L/g
s/\b(excl|frac|ufrac)_auth_agreeL/\1_auth_agree_L/g
# auth_both_valid
s/\bauth_both_valid\b/auth_both_valid_discrete/g
s/\bauth_both_frac_valid\b/auth_both_frac_valid_discrete/g
# gen_heap_ctx and proph_map_ctx
s/\bgen_heap_ctx\b/gen_heap_interp/g
s/\bproph_map_ctx\b/proph_map_interp/g
Ralf Jung
committed
# other gen_heap changes
s/\bmapsto_mapsto_ne\b/mapsto_frac_ne/g
# remove Ts in algebra
s/\bofeT\b/ofe/g
s/\bOfeT\b/Ofe/g
s/\bcmraT\b/cmra/g
s/\bCmraT\b/Cmra/g
# _op/valid/core lemmas
s/\b((coPset|gset)_op)_union\b/\1/g
s/\b((coPset|gset)_core)_self\b/\1/g
s/\b(gmultiset_op)_disj_union\b/\1/g
s/\b(gmultiset_core)_empty\b/\1/g
s/\b(nat_op)_plus\b/\1/g
s/\b(max_nat_op)_max\b/\1/g
Robbert Krebbers
committed
# equiv_spec
s/\bequiv_entails\b/equiv_entails_1_1/g
s/\bequiv_entails_sym\b/equiv_entails_1_2/g
s/\bequiv_spec\b/equiv_entails/g
This release does not have any outstanding highlights, but contains a large
* `heap_lang` now supports deallocation as well as better reasoning about
"invariant locations" (locations that perpetually satisfy some Coq-level
invariant).
* Invariants (`inv N P`) are more flexible, now also supporting splitting
and merging of invariants with respect to separating conjunction.
* Performance of the proofmode for BIs constructed on top of other BIs (e.g.,
`monPred`) was greatly improved, leading to up to 70% speedup in individual
files. As part of this refactoring, the proofmode can now also be instantiated
with entirely "logical" notion of BIs that do not have a (non-trivial) metric
structure, and still support reasoning about ▷.
* The proof mode now provides experimental support for naming pure facts in
intro patterns. See
[iris/string-ident](https://gitlab.mpi-sws.org/iris/string-ident) for details.
* Iris now provides official ASCII notation. We still recommend using the
Unicode notation for better consistency and interoperability with other Iris
libraries, but provide ASCII notation for when Unicode is not an option.
* We removed several coercions, fixing "ambiguous coercion path" warnings and
solving some readability issues.
* Coq 8.10, 8.11, and 8.12 are newly supported by this release, and Coq 8.7 and
8.8 are no longer supported.
Further details are given in the changelog below. We always first list the
potentially breaking changes, then (some of) the additions.
This release of Iris received contributions by Abel Nieto, Amin Timany, Dan
Frumin, Derek Dreyer, Dmitry Khalanskiy, Gregory Malecha, Jacques-Henri Jourdan,
Jonas Kastberg, Jules Jacobs, Matthieu Sozeau, Maxime Dénès, Michael Sammler,
Paolo G. Giarrusso, Ralf Jung, Robbert Krebbers, Simon Friis Vindum, Simon
Spies, and Tej Chajed. Thanks a lot to everyone involved!
**Changes in `heap_lang`:**
* Remove global `Open Scope Z_scope` from `heap_lang.lang`, and leave it up to
reverse dependencies if they want to `Open Scope Z_scope` or not.
* Fix all binary operators performing pointer arithmetic (instead of just the
dedicated `OffsetOp` operator doing that).
* Rename `heap_lang.lifting` to `heap_lang.primitive_laws`. There now also
exists `heap_lang.derived_laws`.
* Make lemma names for `fill` more consistent
- Use the `_inv` suffix for the the backwards directions:
`reducible_fill` → `reducible_fill_inv`,
`reducible_no_obs_fill` → `reducible_no_obs_fill_inv`,
`not_stuck_fill` → `not_stuck_fill_inv`.
- Use the non-`_inv` names (that freed up) for the forwards directions:
`reducible_fill`, `reducible_no_obs_fill`, `irreducible_fill_inv`.
* Remove namespace `N` from `is_lock`.
* Add support for deallocation of locations via the `Free` operation.
* Add a fraction to the heap_lang `array` assertion.
* Add `lib.array` module for deallocating, copying and cloning arrays.
* Add TWP (total weakest-pre) lemmas for arrays.
* Add a library for "invariant locations": heap locations that will not be
deallocated (i.e., they are GC-managed) and satisfy some pure, Coq-level
invariant. See `iris.base_logic.lib.gen_inv_heap` for details.
* Add the ghost state for "invariant locations" to `heapG`. This affects the
statement of `heap_adequacy`, which is now responsible for initializing the
"invariant locations" invariant.
* Add lemma `mapsto_mapsto_ne : ¬ ✓(q1 + q2)%Qp → l1 ↦{q1} v1 -∗ l2 ↦{q2} v2 -∗ ⌜l1 ≠ l2⌝`.
* Add lemma `is_lock_iff` and show that `is_lock` is contractive.
**Changes in `program_logic`:**
* In the axiomatization of ectx languages, replace the axiom of positivity of
context composition with an axiom that says if `fill K e` takes a head step,
then either `K` is the empty evaluation context or `e` is a value.
**Changes in the logic (`base_logic`, `bi`):**
* Rename some accessor-style lemmas to consistently use the suffix `_acc`
`inv_open` → `inv_acc`, `inv_open_strong` → `inv_acc_strong`,
`inv_open_timeless` → `inv_acc_timeless`, `na_inv_open` → `na_inv_acc`,
`cinv_open` → `cinv_acc`, `cinv_open_strong` → `cinv_acc_strong`,
`auth_open` → `auth_acc`, `sts_open` → `sts_acc`.
To make this work, also rename `inv_acc` → `inv_alter`.
(Most developments should be unaffected as the typical way to invoke these
lemmas is through `iInv`, and that does not change.)
* Change `inv_iff`, `cinv_iff` and `na_inv_iff` to make order of arguments
consistent and more convenient for `iApply`. They are now of the form
`inv N P -∗ ▷ □ (P ↔ Q) -∗ inv N Q` and (similar for `na_inv_iff` and
`cinv_iff`), following e.g., `inv_alter` and `wp_wand`.
* Rename `inv_sep_1` → `inv_split_1`, `inv_sep_2` → `inv_split_2`, and
`inv_sep` → `inv_split` to be consistent with the naming convention in boxes.
* Update the strong variant of the accessor lemma for cancellable invariants to
match that of regular invariants, where you can pick the mask at a later time.
(The other part that makes it strong is that you get back the token for the
invariant immediately, not just when the invariant is closed again.)
* Rename `iProp`/`iPreProp` to `iPropO`/`iPrePropO` since they are `ofeT`s.
Introduce `iProp` for the `Type` carrier of `iPropO`.
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
* Flatten the BI hierarchy by merging the `bi` and `sbi` canonical structures.
This gives significant performance benefits on developments that construct BIs
from BIs (e.g., use `monPred`). For, example it gives a performance gain of 37%
overall on lambdarust-weak, with improvements for individual files up to 72%,
see Iris issue #303. The concrete changes are as follows:
+ The `sbi` canonical structure has been removed.
+ The `bi` canonical structure contains the later modality. It does not
require the later modality to be contractive or to satisfy the Löb rule, so
we provide a smart constructor `bi_later_mixin_id` to get the later axioms
"for free" if later is defined to be the identity function.
+ There is a separate class `BiLöb`, and a "free" instance of that class if
the later modality is contractive. A `BiLöb` instance is required for the
`iLöb` tactic, and for timeless instances of implication and wand.
+ There is a separate type class `BiInternalEq` for BIs with a notion of
internal equality (internal equality was part of `sbi`). An instance of this
class is needed for the `iRewrite` tactic, and the various lemmas about
internal equality.
+ The class `SbiEmbed` has been removed and been replaced by classes
`BiEmbedLater` and `BiEmbedInternalEq`.
+ The class `BiPlainly` has been generalized to BIs without internal equality.
As a consequence, there is a separate class `BiPropExt` for BIs with
propositional extensionality (i.e., `■ (P ∗-∗ Q) ⊢ P ≡ Q`).
+ The class `BiEmbedPlainly` is a bi-entailment (i.e., `⎡■ P⎤ ⊣⊢ ■ ⎡P⎤`
instead of `■ ⎡P⎤ ⊢ ⎡■ P⎤`) as it has been generalized to BIs without a
internal equality. In the past, the left-to-right direction was obtained for