When using `set_option tactic.skipAssignedInstances false`, `simp` and
`rw` will synthesize instance implicit arguments even if they have
assigned by unification. If the synthesized argument does not match the
assigned one the rewrite is not performed. This option has been added
for backward compatibility.
```
example (a : Nat) :
(((a + (2 ^ 64 - 1)) % 2 ^ 64 + 1) * 8 - 1 - (a + (2 ^ 64 - 1)) % 2 ^ 64 * 8 + 1) = 8 := by
omega
```
used to time out, and now is fast.
(We will probably make separate changes later so the defeq checks would
be fast in any case here.)
The `decide` tactic produces error messages that users find to be
obscure. Now:
1. If the `Decidable` instance reduces to `isFalse`, it reports that
`decide` failed because the proposition is false.
2. If the `Decidable` instance fails to reduce, it explains what
proposition it failed for, and it shows the reduced `Decidable` instance
rather than the `Decidable.decide` expression. That expression tends to
be less useful since it shows the unreduced `Decidable` argument (plus
it's a lot longer!)
Examples:
```lean
example : 1 ≠ 1 := by decide
/-
tactic 'decide' proved that the proposition
1 ≠ 1
is false
-/
opaque unknownProp : Prop
open scoped Classical in
example : unknownProp := by decide
/-
tactic 'decide' failed for proposition
unknownProp
since its 'Decidable' instance reduced to
Classical.choice ⋯
rather than to the 'isTrue' constructor.
-/
```
When reporting the error, `decide` only shows the whnf of the
`Decidable` instance. In the future we could consider having it reduce
all decidable instances present in the term, which can help with
determining the cause of failure (this was explored in
8cede580690faa5ce18683f168838b08b372bacb).
The elaboration function `Lean.Meta.coerceMonadLift?` inserts these
coercion helper functions into a term and tries to unfolded them with
`expandCoe`, but because that function only unfolds up to
reducible-and-instance transparency, these functions were not being
unfolded. The fix here is to give them the `@[reducible]` attribute.
with this, more functions will be proven terminating automatically,
namely those where after `simp_wf`, lexicographic order handling,
possibly `subst_vars` the remaining goal can be solved by `omega`.
Note that `simp_wf` already does simplification of the goal, so
this adds `omega`, not `(try simp) <;> omega` here.
There are certainly cases where `(try simp) <;> omega` will solve more
goals (e.g. due to the `subst_vars` in `decreasing_with`), and
`(try simp at *) <;> omega` even more. This PR errs on the side of
taking
smaller steps.
Just appending `<;> omega` to the existing
`simp (config := { arith := true, failIfUnchanged := false })` call
doesn’t work nicely, as that leaves forms like `Nat.sub` in the goal
that
`omega` does not seem to recognize.
This does *not* remove any of the existing ad-hoc `decreasing_trivial`
rules based on `apply` and `assumption`, to not regress over the status
quo (these rules may apply in cases where `omega` wouldn't “see”
everything, but `apply` due to defeq works).
Additionally, just extending makes bootstrapping easier; early in `Init`
where
`omega` does not work yet these other tactics can still be used.
(Using a single `omega`-based tactic was tried in #3478 but isn’t quite
possible yet, and will be postponed until we have better automation
including forward reasoning.)
with this, hopefully more obvious array accesses will be handled
automatically.
Just like #3503, this PR does not investiate which of the exitsting
tactics in `get_elem_tactic_trivial` are subsumed now and could be
dropped without (too much) breakage.
Before, app unexpanders would only be applied to entire applications.
However, some notations produce functions, and these functions can be
given additional arguments. The solution so far has been to write app
unexpanders so that they can take an arbitrary number of additional
arguments. However, as reported in [this Zulip
thread](https://leanprover.zulipchat.com/#narrow/stream/270676-lean4/topic/pretty.20printer.20bug/near/420662236),
this leads to misleading hover information in the Infoview. For example,
while `HAdd.hAdd f g 1` pretty prints as `(f + g) 1`, hovering over `f +
g` shows `f`. There is no way to fix the situation from within an app
unexpander; the expression position for `HAdd.hAdd f g` is absent, and
app unexpanders cannot register TermInfo.
This commit changes the app delaborator to try running app unexpanders
on every prefix of an application, from longest to shortest prefix. For
efficiency, it is careful to only try this when app delaborators do in
fact exist for the head constant, and it also ensures arguments are only
delaborated once. Then, in `(f + g) 1`, the `f + g` gets TermInfo
registered for that subexpression, making it properly hoverable.
The app delaborator is also refactored, and there are some bug fixes:
- app unexpanders only run when `pp.explicit` is false
- trailing parameters in under-applied applications are now only
considered up to reducible & instance transparency, which lets, for
example, optional arguments for `IO`-valued functions to be omitted.
(`IO` is a reader monad, so it's hiding a pi type)
- app unexpanders will no longer run for delaborators that use
`withOverApp`
- auto parameters now always pretty print, since we are not verifying
that the provided argument equals the result of evaluating the tactic
Furthermore, the `notation` command has been modified to generate an app
unexpander that relies on the app delaborator's new behavior.
The change to app unexpanders is reverse-compatible, but it's
recommended to update `@[app_unexpander]`s in downstream projects so
that they no longer handle overapplication themselves.
This PR addresses several performance issues in the auto-completion
implementation. It also fixes a number of smaller bugs related to
auto-completion.
In a file with `import Mathlib`, the performance of various kinds of
completions has improved as follows:
- Completing `C`: 49000ms -> 1400ms
- Completing `Cat`: 14300ms -> 1000ms
- Completing `x.` for `x : Nat`: 3700ms -> 220ms
- Completing `.` for an expected type of `Nat`: 11000ms -> 180ms
The following bugs have been fixed as well:
- VS Code never used our custom completion order. Now, the server fuzzy
completion score decides the order that completions appear in.
- Dot auto-completion for private types did not work at all. It does
now.
- Completing `.<identifier>` (where the expected type is used to infer
the namespace) did not filter by the expected type and instead displayed
all matching constants in the respective namespace. Now, it uses the
expected type for filtering. Note that this is not perfect because
sub-namespaces are technically correct completions as well (e.g.
`.Foo.foobar`). Implementing this is future work.
- Completing `.` was often not possible at all. Now, as long as the `.`
is not used in a bracket (where it may be used for the anonymous lambda
feature, e.g. `(. + 1)`), it triggers the correct completion.
- Fixes#3228.
- The auto-completion in `#check` commands would always try to complete
identifiers using the full declaration name (including namespaces) if it
could be resolved. Now it simply uses the identifier itself in case
users want to complete this identifier to another identifier.
## Details
Regarding completion performance, I have more ideas on how to improve it
further in the future.
Other changes:
- The feature that completions with a matching expected type are sorted
to the top of the server-side ordering was removed. This was never
enabled in VS Code because it would use its own completion item order
and when testing it I found it to be more confusing than useful.
- In the server-side ordering, we would always display keywords at the
top of the list. They are now displayed according to their fuzzy match
score as well.
The following approaches have been used to improve performance:
- Pretty-printing the type for every single completion made up a
significant amount of the time needed to compute the completions. We now
do not pretty-print the type for every single completion that is offered
to the user anymore. Instead, the language server now supports
`completionItem/resolve` requests to compute the type lazily when the
user selects a completion item.
- Note that we need to keep the amount of properties that we compute in
a resolve request to a minimum. When the server receives the resolve
request, the document state may have changed from the state it was in
when the initial auto-completion request was received. LSP doesn't tell
us when it will stop sending resolve requests, so we cannot keep this
state around, as we would have to keep it around forever.
LSP's solution for this dilemma is to have servers send all the state
they need to compute a response to a resolve request to the client as
part of the initial auto completion response (which then sends it back
as part of the resolve request), but this is clearly infeasible for all
real language servers where the amount of state needed to resolve a
request is massive.
This means that the only practical solution is to use the current state
to compute a response to the resolve request, which may yield an
incorrect result. This scenario can especially occur when using
LiveShare where the document is edited by another person while cycling
through available completions.
- Request handlers can now specify a "header caching handler" that is
called after elaborating the header of a file. Request handlers can use
this caching handler to compute caches for information stored in the
header. The auto-completion uses this to pre-compute non-blacklisted
imported declarations, which in turn allow us to iterate only over
non-blacklisted imported declarations where we would before iterate over
all declarations in the environment. This is significant because
blacklisted declarations make up about 4/5 of all declarations.
- Dot completion now looks up names modulo private prefixes to figure
out whether a declaration is in the namespace of the type to the left of
the dot instead of first stripping the private prefix from the name and
then comparing it. This has the benefit that we do not need to scan the
full name in most cases.
This PR also adds a couple of regression tests for fixed bugs, but *no
benchmarks*. We will add these in the future when we add proper support
for benchmarking server interaction sessions to our benchmarking
architecture.
All tests that were broken by producing different completion output
(empty `detail` field, added `sortText?` and `data?` fields) have been
manually checked by me to be still correct before replacing their
expected output.
The current `ToExpr Int` instance produces `@Int.ofNat (@OfNat.ofNat Nat
i ...)` for nonnegative `i` and `@Int.negSucc (@OfNat.ofNat Nat (-i+1)
...)` for negative `i`.
However it should be producing `@OfNat.ofNat Int i ...` for nonnegative
`i`, and `@Neg.neg ... (@OfNat.ofNat Int (-i) ...)` for negative `i`.
This is very helpful when dealing with bitvectors, where a case analysis
on the bitwidth leaves one with hypotheses of the form `x<2^(Nat.succ
w)`.
Design decisions I am unsure about:
- Is creating a helper `succ?` the correct way to match on the exponent
`e+1`?
- I'm not certain why the prior call to `Int.ofNat_pow` also checked
that the exponent was a ground natural. I removed this, since we now
explicitly handle cases where the exponent is a term of the form `e+1`.
---------
Co-authored-by: Scott Morrison <scott.morrison@gmail.com>
Co-authored-by: Joe Hendrix <joe@lean-fro.org>
Co-authored-by: Alex Keizer <alex@keizer.dev>
This PR is an effort to improve reasoning at the Nat level about
bitvectors and reduce of Fin and Nat.
It slightly tightens some proofs, but is generally aimed at reducing
inconsistencies between definitions at the Nat and Fin types in favor of
more consistently using Nat operations.
This ports leanprover/std4#664 to Lean core.
Here was the rational I provided in the discussion for
leanprover/std4#664:
It's mostly about consistency. If we use the same types and style in
definitions and proofs, there is less surprise when unfolding or
otherwise using definitions. We use some Nat based operations that
haven't been extended to Fin such as the bitwise operations, and I don't
want to pay the overhead of introducing a Fin version of every Bitvector
operation.
So this basically means Nat is preferred.
One argument potentially in favor of Fin is that we could reuse results
proven there, but that doesn't really seem to be the case so far.
A second argument is that we want to simplify expression to use more
canonical forms and we currently can pretty-print those operations
better using ofNat than ofFin. We could define the notations using ofFin
of course though, but that's additional operators that will show up in
expressions.
Adds a simple error-recovery mechanism to Lean's parser, similar to
those used in other combinator parsing libraries.
Lean itself isn't very amenable to error recovery with this mechanism,
as it requires global knowledge of the grammar in question to write
recovery rules that don't break backtracking or `<|>`. I only found a
few opportunities.
But for DSLs, this is really important. In particular, Verso parse
errors interacted very badly with Lean parse errors in a way that
required frequent "restart file" commands, but this mechanism allows me
to both recover from Verso parse errors and to have Lean skip the rest
of the file rather than repeatedly trying to parse it as Lean commands.
This is a quite substantial tactic.
It also includes the infamour `NatCast` typeclass (which I've equipped
with a module-doc). I wasn't at all sure where that should live, so it
is currently randomly in `Lean/Elan/Tactic/NatCast.lean`: presumably if
we're doing this it will go somewhere in `Init`.
---------
Co-authored-by: Leonardo de Moura <leomoura@amazon.com>
