This PR redefines `String` to be the type of byte arrays `b` for which
`b.IsValidUtf8`.
This moves the data model of strings much closer to the actual data
representation at runtime.
In the near future, we will
- provide variants of `String.Pos` and `Substring` that only allow for
valid positions
- redefine all `String` functions to be much closer to their C++
implementations
In the near-to-medium future we will then provide comprehensive
verification of `String` based on these refactors.
This PR adds support the Count Trailing Zeros operation `BitVec.ctz` to
the bitvector library and to `bv_decide`, relying on the existing `clz`
circuit. We also build some theory around `BitVec.ctz` (analogous to the
theory existing for `BitVec.clz`) and introduce lemmas
`BitVec.[ctz_eq_reverse_clz, clz_eq_reverse_ctz, ctz_lt_iff_ne_zero,
getLsbD_false_of_lt_ctz, getLsbD_true_ctz_of_ne_zero,
two_pow_ctz_le_toNat_of_ne_zero, reverse_reverse_eq,
reverse_eq_zero_iff]`.
`ctz` operation is common in numerous compiler intrinsics (see
[here](https://clang.llvm.org/docs/LanguageExtensions.html#intrinsics-support-within-constant-expressions))
and architectures (see
[here](https://en.wikipedia.org/wiki/Find_first_set)).
---------
Co-authored-by: Siddharth <siddu.druid@gmail.com>
This PR enables the new E-matching pattern inference heuristic for
`grind`, implemented in PR #10422.
**Important**: Users can still use the old pattern inference heuristic
by setting:
```lean
set_option backward.grind.inferPattern true
```
In PR #10422, we introduced the new modifier `@[grind!]` for enabling
the minimal indexable subexpression condition. This option can now also
be set in `grind` parameters. Example:
```lean
opaque f : Nat → Nat
opaque fInv : Nat → Nat
axiom fInv_f : fInv (f x) = x
/-- trace: [grind.ematch.pattern] fInv_f: [f #0] -/
#guard_msgs in
set_option trace.grind.ematch.pattern true in
example {x y} : f x = f y → x = y := by
/-
The modifier `!` instructs `grind` to use the minimal indexable subexpression
(i.e., `f x` in this case).
-/
grind [!fInv_f]
```
This PR fixes a few bugs in the `rw` tactic: it could "steal" goals
because they appear in the type of the rewrite, it did not do an occurs
check, and new proof goals would not be synthetic opaque. This PR also
lets the `rfl` tactic assign synthetic opaque metavariables so that it
is equivalent to `exact rfl`.
Implementation note: filtering old vs new is not sufficient. This PR
partially addresses the bug where the rw tactic creates natural
metavariables for each of the goals; now new proof goals are synthetic
opaque.
Metaprogramming API: Instead of `Lean.MVarId.rewrite` prefer
`Lean.Elab.Tactic.elabRewrite` for elaborating rewrite theorems and
applying rewrites to expressions.
Closes#10172
This PR adds range support to`BitVec` and the `UInt*` types. This means
that it is now possible to write, for example, `for i in (1 : UInt8)...5
do`, in order to loop over the values 1, 2, 3 and 4 of type `UInt8`.
This PR adds more lemmas about the `toList` and `toArray` functions on
ranges and iterators. It also renames `Array.mem_toArray` into
`List.mem_toArray`.
This PR is followup to the change in grind pattern heuristics from
#10342, typically resolving the discrepancy by writing out an explicit
`grind_pattern` for the intended pattern. The new behaviour is more
aggressive, because it selects smaller patterns.
This PR completes the review of `@[grind]` annotations without a sigil
(e.g. `=` or `←`), replacing most of them with more specific annotations
or patterns.
---------
Co-authored-by: Leonardo de Moura <leomoura@amazon.com>
This PR moves the definitions and basic facts about `Function.Injective`
and `Function.Surjective` up from Mathlib. We can do a better job of
arguing via injectivity in `grind` if these are available.
This PR updates `@[grind]` annotations which should be `@[grind =]`, for
robustness (and, presumably, in some fraction of cases the existing
heuristic for `@[grind]` is already too liberal).
This PR removes `grind →` annotations that fire too often, unhelpfully.
It would be nice for `grind` to instantiate these lemmas, but only if
they already see `xs ++ ys` and `#[]` in the same equivalence class, not
just as soon as it sees `xs ++ ys`.
In the meantime, let's see what is using these.
This PR prepares for a future reorganization of the import hierarchy so
that `Init.Data.String.Basic` can import `Init.Data.UInt.Bitwise` and
`Init.Data.Array.Lemmas`.
This PR moves `String.utf8EncodeChar` to the prelude to prepare for the
imminent redefinition of `String`.
The definition in the prelude uses modulo and division operations on
natural numbers. In `String.Extra`, a `csimp` lemma is provided, showing
that the new definition is equal to the previous one (which is now
called `utf8EncodeCharFast`) which uses bitwise operations on `UInt8`.
This PR improves the names of definitions and lemmas in the polymorphic
range API. It also introduces a recommended spelling. For example, a
left-closed, right-open range is spelled `Rco` in analogy with Mathlib's
`Ico` intervals.
This PR speeds up auto-completion by a factor of ~3.5x through various
performance improvements in the language server. On one machine, with
`import Mathlib`, completing `i` used to take 3200ms and now instead
yields a result in 920ms.
Specifically, the following improvements are made:
- The watchdog process no longer de-serializes and re-serializes most
messages from the file worker before passing them on to the user - a
fast partial de-serialization procedure is now used to determine whether
the message needs to be de-serialized in full or not.
- `escapePart` is optimized to perform better on ASCII strings that do
not need escaping.
- `Json.compress` is optimized to allocate fewer objects.
- A faster JSON compression specifically for completion responses is
implemented that skips allocating `Json` altogether.
- The JSON compression has been moved to the task where we convert a
request response to `Json` so that converting to a string won't block
the output task of the FileWorker and so the `Json` value is not marked
as multi-threaded when we compress is, which drastically increases the
cost of reference-counting.
- The JSON representation of the `data?` field of each completion item
is optimized.
- Both the completion kind and the set of completion tags for each
imported completion item is now cached.
- The filtering of duplicate completion items is optimized.
Other adjustments:
- `LT UInt8` and `LE UInt8` are moved to Prelude so that they can be
used in `Init.Meta` for the name part escaping fast path.
- `Array.usize` is exposed since it was marked as `@[simp]`.
This PR fixes a bug in the `LinearOrderPackage.ofOrd` factory. If there
is a `LawfulEqOrd` instance available, it should automatically use it
instead of requiring the user to provide the `eq_of_compare` argument to
the factory. The PR also solves a hygiene-related problem making the
factories fail when `Std` is not open.
This PR adds some test cases for `grind` working with `Fin`. There are
many still failing tests in `tests/lean/grind/grind_fin.lean` which I'm
intending to triage and work on.
This PR is the result of analyzing the elaborator performance regression
introduced by #10005. It makes the `workspaceSymboldNewRanges` and
`iterators` benchmarks less noisy. It also replaces some range-related
instances for `Nat` with shortcuts to the general-purpose instances.
This is a trade-off between the ergonomics and the synthesis cost of
having general-purpose instances.
This PR adds the inverse of a dyadic rational, at a given precision, and
characterising lemmas. Also cleans up various parts of the `Int.DivMod`
and `Rat` APIs, and proves some characterising lemmas about
`Rat.toDyadic`.
---------
Co-authored-by: Rob23oba <152706811+Rob23oba@users.noreply.github.com>
This PR adds “non-branching case statements”: For each inductive
constructor `T.con` this adds a function `T.con.with` that is similar
`T.casesOn`, but has only one arm (the one for `con`), and an additional
`t.toCtorIdx = 12` assumption.
For example:
```lean
inductive Vec (α : Type) : Nat → Type where
| nil : Vec α 0
| cons {n} : α → Vec α n → Vec α (n + 1)
/--
info: @[reducible] protected def Vec.cons.elim.{u} : {α : Type} →
{motive : (a : Nat) → Vec α a → Sort u} →
{a : Nat} →
(t : Vec α a) →
t.ctorIdx = 1 → ({n : Nat} → (a : α) → (a_1 : Vec α n) → motive (n + 1) (Vec.cons a a_1)) → motive a t
-/
#guard_msgs in
#print sig Vec.cons.elim
```
This is a building block for non-quadratic implementations of `BEq` and
`DecidableEq` etc.
Builds on top of #9951.
The compiled code for a these functions could presumably, without
branching on the inductive value, directly access the fields. Achieving
this optimization (and achieving it without a quadratic compilation
cost) is not in scope for this PR.
This PR implements the fast circuit for overflow detection in unsigned
multiplication used by Bitwuzla and proposed in:
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=987767
The theorem is based on three definitions:
* `uppcRec`: the unsigned parallel prefix circuit for the bits until a
certain `i`
* `aandRec`: the conjunction between the parallel prefix circuit at of
the first operand until a certain `i` and the `i`-th bit in the second
operand
* `resRec`: the preliminary overflow flag computed with these two
definitions
To establish the correspondence between these definitiions and their
meaning in `Nat`, we rely on `clz` and `clzAuxRec` definitions.
Therefore, this PR contains the `clz`- and `clzAuxRec`-related
infrastructure that was necessary to get the proofs through.
An additional change this PR contains is the moving of `### Count
leading zeros` section in `BitVec.Lemmas` downwards. In fact, some of
the proofs I wrote required introducing `Bitvec.toNat_lt_iff` and
`BitVec.le_toNat_iff` which I believe should live in the `Inequalities`
section. Therefore, to put these in the appropriate section, I decided
to move the whole `clz` section downwards (while it's small and
relatively self contained. Specifically, the theorems I moved are:
`clzAuxRec_zero`, `clzAuxRec_succ`, `clzAuxRec_eq_clzAuxRec_of_le`,
`clzAuxRec_eq_clzAuxRec_of_getLsbD_false`.
The fast circuit is not yet the default one in the bitblaster, as it's
performance is not yet competitive due to some missing rewrites that
bitwuzla supports but are not in Lean yet.
co-authored-by: @bollu
---------
Co-authored-by: Tobias Grosser <tobias@grosser.es>
This PR defines the dyadic rationals, showing they are an ordered ring
embedding into the rationals. We will use this for future interval
arithmetic tactics.
Many thanks to @Rob23oba, who did most of the implementation work here.
---------
Co-authored-by: Rob23oba <robin.arnez@web.de>
This PR generates `.ctorIdx` functions for all inductive types, not just
enumeration types. This can be a building block for other constructions
(`BEq`, `noConfusion`) that are size-efficient even for large
inductives.
It also renames it from `.toCtorIdx` to `.ctorIdx`, which is the more
idiomatic naming.
The old name exists as an alias, with a deprecation attribute to be
added after the next
stage0 update.
These functions can arguably compiled down to a rather efficient tag
lookup, rather than a `case` statement. This is future work (but
hopefully near future).
For a fair number of basic types the compiler is not able to compile a
function using `casesOn` until further definitions have been defined.
This therefore (ab)uses the `genInjectivity` flag and
`gen_injective_theorems%` command to also control the generation of this
construct.
For (slightly) more efficient kernel reduction one could use `.rec`
rather than `.casesOn`. I did not do that yet, also because it
complicates compilation.
This PR upstreams lemmas about `Rat` from `Mathlib.Data.Rat.Defs` and
`Mathlib.Algebra.Order.Ring.Unbundled.Rat`, specifically enough to get
`Lean.Grind.Field Rat` and `Lean.Grind.OrderedRing Rat`. In addition to
the lemmas, instances for `Inv Rat`, `Pow Rat Nat` and `Pow Rat Int`
have been upstreamed.
---------
Co-authored-by: Kim Morrison <kim@tqft.net>
This PR contains lemmas about `Int` (minor amendments for BitVec and
Nat) that are being used in preparing the dyadics. This is all work of
@Rob23oba, which I'm pulling out of #9993 early to keep that one
manageable.
This PR improves support for `a^n` in `grind cutsat`. For example, if
`cutsat` discovers that `a` and `b` are equal to numerals, it now
propagates the equality. This PR is similar to #9996, but `a^b`.
Example:
```lean
example (n : Nat) : n = 2 → 2 ^ (n+1) = 8 := by
grind
```
With #10022, it also improves the support for `BitVec n` when `n` is not
numeral. Example:
```lean
example {n m : Nat} (x : BitVec n)
: 2 ≤ n → n ≤ m → m = 2 → x = 0 ∨ x = 1 ∨ x = 2 ∨ x = 3 := by
grind
```
This PR reverts parts of #10005 that surprisingly turned out to cause a
performance regression in the benchmarks. The slowdown seems to be
related to elaboration, not inefficiencies in the generated code. This
is just a quick fix. I will take a closer look in a week.
This PR adds a stop position field to parser input contexts, allowing
the parser to be instructed to stop parsing prior to the end of a file.
This is step 1, prior to a stage0 update, to make run-time data
structures sufficiently compatible to avoid segfaults. After the update,
the actual code to stop parsing can be merged.
