This PR surfaces kernel diagnostics even in `example`.
The problem was that the kernel checking happens asynchronously. We
cannot use `reportDiag` in `addDecl`, which spawns that task, due to the
module hierarchy. For non `example`-declaration, `reportDiag` is called
somewhere else later, but for `example`, the `withoutModifyingEnv` in
`elabMutualDef` hid the kernel diagnostics. (But only the kernel
diagnostics; they are in the `Environment`, while the others are in the
`State`).
I also observed that the `reportDiag` in `elabAsync` (but not in
`elabSync`) duplicated the reporting, so without `elab.Async true` you
get the message twice. To fix this, `reportDiag` now resets the
diagnostics. This should avoid reporting counts twice in general (at
least within a linear use of the state).
---------
Co-authored-by: Sebastian Ullrich <sebasti@nullri.ch>
This PR makes `simp` consult its own cache more often, to avoid
replicating work.
Before, the simp cache was checked upon entry of `simpImpl` only, which
then calls `simpLoop`, which recursively iterates the `pre`-lemmas,
without checking the cache again.
Now, `simpLoop` itself checks the cache. This seems more principled,
given that `simpLoop` is actually putting entries into the cache for
each of its calls, so it’s more uniform if it checks the cache itself.
This avoids repeated rewrites. For example given
```
theorem ab : a = b := testSorry
theorem bc : b = c := testSorry
example (h : P c) : P b ∧ P a := by simp [ab, bc, h]
```
simp would rewrite `b ==> c` twice (once as part of `b ==> c` and then
again as part of `a ==> b ==> c`). And it’d be order dependent: With
```
example (h : P c) : P a ∧ P b := by simp [ab, bc, h]
```
the `a ==> b ==> c` chain would insert `b ==> c` into the cache, and
picked up by `simpImpl` when rewriting `P b`.
With this change, `b ==> c` is performed only once in both examples.
Instruction counts on stdlib and mathlib both show a mild improvement
across the board (0.5%), with individual modules improving by up to 4%
in stdlib and even more in mathlib.
(This does not check the cache before applying `post`, which explains
where there are still some repeated rewrites in the trace logs. But I’m
less sure about inserting a cache check here and so I am treading
carefully here. It’s also going to be at most one `post` application
that’s duplicated, because if `post` returns `.visit`, we go back to
`pre` and thus a cache check.)
This PR makes `#guard_msgs` to treat `trace` messages separate from
`info`, `warning` and `error`. It also introduce the ability to say
`#guard_msgs (pass info`, like `(drop info)` so far, and also adds
`(check info)` as the explicit form of `(info)`, for completeness.
Fixes#8266
This PR makes functions defined by well-founded recursion use an
`opaque` well-founded proof by default. This reliably prevents kernel
reduction of such definitions and proofs, which tends to be
prohibitively slow (fixes#2171), and which regularly causes
hard-to-debug kernel type-checking failures. This changes renders
`unseal` ineffective for such definitions. To avoid the opaque proof,
annotate the function definition with `@[semireducible]`.
This PR enables the elaboration of theorem bodies, i.e. proofs, to
happen in parallel to each other as well as to other elaboration tasks.
Specifically, to be eligible for parallel proof elaboration,
* the theorem must not be in a `mutual` block
* `deprecated.oldSectionVars` must not be set
* `Elab.async` must be set (currently defaults to `true` in the language
server, `false` on the cmdline)
To be activated for downstream projects (i.e. in stage 1) pending
further Mathlib validation.
This PR fixes the caching infrastructure for `whnf` and `isDefEq`,
ensuring the cache accounts for all relevant configuration flags. It
also cleans up the `WHNF.lean` module and improves the configuration of
`whnf`.
Sebastian mentioned that the use of the kernel defeq was to work around
a performance issue that was fixed since. Let's see if we can do
without.
This is also a semantic change: Ground terms (no free vars, no mvars)
are reduced at
“all” transparency even if the the transparency setting is default. This
was the case
even before 03f6b87647 switched to the
kernel defeq
checking for performance. It seems that this is rather surprising
behavior from the user
point of view. The fallout on batteries and mathlib is rather limited,
only a few
`rfl` proofs seem to have (inadvertently or not) have relied on this.
The speedcenter reports no significant regressions on core or mathlib.
Currently, the messages in the diagnostic summaries are created by
appending interpolated strings. We wrap these in `.trace`'s, and the
results are better formatted when expanding child nodes in the info
view. Particularly, the latter diagnostic summaries remain on their own
lines flush to the left instead of on the same line directly adjacent to
the last child node.
we keep running into examples where working with well-founded recursion
is slow because defeq checks (which are all over the place, including
failing ones that are back-tracked) unfold well-founded definitions.
The definition of a function defined by well-founded recursion should be
an implementation detail that should only be peeked inside by the
equation generator and the functional induction generator.
We now mark the mutual recursive function as irreducible (if the user
did not
set a flag explicitly), and use `withAtLeastTransparency .all` when
producing
the equations.
Proofs can be fixed by using rewriting, or – a bit blunt, but nice for
adjusting
existing proofs – using `unseal` (a.k.a. `attribute [local
semireducible]`).
Mathlib performance does not change a whole lot:
http://speed.lean-fro.org/mathlib4/compare/08b82265-75db-4a28-b12b-08751b9ad04a/to/16f46d5e-28b1-41c4-a107-a6f6594841f8
Build instructions -0.126 %, four modules with significant instructions
decrease.
To reduce impact, these definitions were changed:
* `Nat.mod`, to make `1 % n` reduce definitionally, so that `1` as a
`Fin 2` literal
works nicely. Theorems with larger `Fin` literals tend to need a `unseal
Nat.modCore`
https://github.com/leanprover/lean4/pull/4098
* `List.ofFn` rewritten to be structurally recursive and not go via
`Array.ofFn`:
https://github.com/leanprover-community/batteries/pull/784
Alternative designs explored were
* Making `WellFounded.fix` irreducible.
One benefit is that recursive functions with equal definitions (possibly
after
instantiating fixed parameters) are defeq; this is used in mathlib to
relate
[`OrdinalApprox.gfpApprox`](https://leanprover-community.github.io/mathlib4_docs/Mathlib/SetTheory/Ordinal/FixedPointApproximants.html#OrdinalApprox.gfpApprox)
with `.lfpApprox`.
But the downside is that one cannot use `unseal` in a
targeted way, being explicit in which recursive function needs to be
reducible here.
And in cases where Lean does unwanted unfolding, we’d still unfold the
recursive
definition once to expose `WellFounded.fix`, leading to large terms for
often no good
reason.
* Defining `WellFounded.fix` to unroll defintionally once before hitting
a irreducible
`WellFounded.fixF`. This was explored in #4002. It shares most of the
ups and downs
with the previous variant, with the additional neat benefit that
function calls that
do not lead to recursive cases (e.g. a `[]` base case) reduce nicely.
This means that
the majority of existing `rfl` proofs continue to work.
Issue #4051, which demonstrates how badly things can go if wf recursive
functions can be
unrolled, showed that making the recursive function irreducible there
leads to noticeably
faster elaboration than making `WellFounded.fix` irreducible; this is
good evidence that
the present PR is the way to go.
This fixes https://github.com/leanprover/lean4/issues/3988
---------
Co-authored-by: Leonardo de Moura <leomoura@amazon.com>
This change
* moves `termination_by` and `decreasing_by` next to the function they
apply to
* simplify the syntax of `termination_by`
* apply the `decreasing_by` goal to all goals at once, for better
interactive use.
See the section in `RELEASES.md` for more details and migration advise.
This is a hard breaking change, requiring developers to touch every
`termination_by` in their code base. We decided to still do it as a
hard-breaking change, because supporting both old and new syntax at the
same time would be non-trivial, and not save that much. Moreover, this
requires changes to some metaprograms that developers might have
written, and supporting both syntaxes at the same time would make
_their_ migration harder.
