Due to nested recursion, we do two passes of `getRecArgInfo`: One on
each argument in isolation, to see which inductive types are around
(e.g. `Tree` and `List`), and
then we later refine/replace this result with the data for the nested
type former (the implicit `ListTree`).
If we have nested recursion through a non-recursive data type like
`Array` or `Prod` then arguemnts of these types should survive the first
phase, so that we can still use them when looking for, say, `Array
Tree`.
This was helpfully reported by @arthur-adjedj.
the support for mutual structural recursion (new since #4575) is
extended so that Lean tries to infer it even without annotations.
* The error message when termination checking fails looks quite
different now. Maybe a bit better, maybe with more room for
improvements.
* If there are too many combinations (with an arbitrary cut-off) for a
given argument type, it will just give up and ask the user to use
`termination_by structural`.
* It is now legal to specify `termination_by structural` on not
necessarily all functions of a clique; this simply restricts the
combinations of arguments that Lean considers.
---------
Co-authored-by: Tobias Grosser <tobias@grosser.es>
This adds support for mutual structural recursive functions.
For now this is opt-in: The functions must have a `termination_by
structural …` annotation (new since #4542) for this to work:
```lean
mutual
inductive A
| self : A → A
| other : B → A
| empty
inductive B
| self : B → B
| other : A → B
| empty
end
mutual
def A.size : A → Nat
| .self a => a.size + 1
| .other b => b.size + 1
| .empty => 0
termination_by structural x => x
def B.size : B → Nat
| .self b => b.size + 1
| .other a => a.size + 1
| .empty => 0
termination_by structural x => x
end
```
The recursive functions don’t have to be in a one-to-one relation to a
set of mutually recursive inductive data types. It is possible to ignore
some of the types:
```lean
def A.self_size : A → Nat
| .self a => a.self_size + 1
| .other _ => 0
| .empty => 0
termination_by structural x => x
```
or have more than one function per argument type:
```lean
def isEven : Nat → Prop
| 0 => True
| n+1 => ¬ isOdd n
termination_by structural x => x
def isOdd : Nat → Prop
| 0 => False
| n+1 => ¬ isEven n
termination_by structural x => x
```
This does not include
* Support for nested inductive data types or nested recursion
* Inferring mutual structural recursion in the absence of
`termination_by`.
* Functional induction principles for these.
* Mutually recursive functions that live in different universes. This
may be possible,
maybe after beefing up the `.below` and `.brecOn` functions; we can look
into this some
other time, maybe when there are concrete use cases.
---------
Co-authored-by: Richard Kiss <him@richardkiss.com>
Co-authored-by: Tobias Grosser <tobias@grosser.es>
this is in preparation for #4542, and extracts from `findRecArg` the
functionality for trying one particular argument.
It also refactors the code a bit. In particular
* It reports errors in the order of the parameters, not the order of
in which they are tried (it tries non-indices first).
* For every argument it will say why it wasn't tried, even if the
reason is quite obviously (fixed prefix, or `Prop`-typed etc.)
Therefore there is some error message churn.
Sets the default value to `pp.fieldNotation.generalized` to `true`.
Updates tests, and fixes some minor flaws in the implementation of the
generalized field notation pretty printer.
Now generalized field notation won't be used for any function that has a
`motive` argument. This is intended to prevent recursors from pretty
printing using it as (1) recursors are more like control flow structures
than actual functions and (2) generalized field notation tends to cause
elaboration problems for recursors.
Note: be sure functions that have an `@[app_unexpander]` use
`@[pp_nodot]` if applicable. For example, `List.toArray` needs
`@[pp_nodot]` to ensure the unexpander prints it using `#[...]`
notation.
this refactor prepares GuessLex to be able to infer more complex
termination arguments.
As a side-effect it fixes an (obscure) bug where `sizeOf` would be
applied to a term of the wrong type and thus a wrong `SizeOf` instance
could be inferred.
This introduces the `ArgsPacker` module and abstraction, to replace the
exising `PackDomain`/`PackMutual` code. The motivation was that we now
have more uses besides `Fix.lean` (`GuessLex` and `FunInd`), and the
code was spread in various places.
The goals are
* consistent function naming withing the the `PSigma` handling, the
`PSum` handling, and the combined interface
* avoid taking a type apart just based on the `PSigma`/`PSum` nesting,
to be robust in case the user happens to be using `PSigma`/`PSum`
somewhere. Therefore, always pass an `arity` or `numFuncs` or `varNames`
around.
* keep all the `PSigma`/`PSum` encoding logic contained within one
module (`ArgsPacker`), and keep that module independent of its users (so
no `EqnInfos` visible here).
* pick good variable names when matching on a packed argument
* the unary function now is either called `fun1._unary` or
`fun1._mutual`, never `fun1._unary._mutual`.
This file has less heavy dependencies than `PackMutual` had, so build
parallelism is improved as well.
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.)
The elaborator is prone to duplicate terms, including recursive calls,
even if the user only wrote a single one. This duplication is wasteful
if we run the tactics on duplicated calls, and confusing in the output
of GuessLex. So prune the list of recursive calls, and remove those
where another call exists that has the same goal and context that is no
more specific.
by showing the matrix of calls and measures, and what we know about that
call (=, <, ≤, ?), e.g.
guessLexFailures.lean:27:0-33:31: error: Could not find a decreasing
measure.
The arguments relate at each recursive call as follows:
(<, ≤, =: relation proved, ? all proofs failed, _: no proof attempted)
x1 x2 x3
1) 29:6-25 = = =
2) 30:6-23 = ? <
3) 31:6-23 < _ _
Please use `termination_by` to specify a decreasing measure
It’s a bit more verbose for mutual functions.
It will use the user-specified argument names for functions written
```
foo (n : Nat) := …
```
but not with pattern matching like
```
foo : Nat → …
| n => …
```
This can be refined later and separately (and maybe right away in
`expandMatchAltsWhereDecls`).
If here is only one plausible measure, there is no point having the
`GuessLex` code see if it
is terminating, running all the tactics, only for the `MkFix` code then
run the tactics again.
So if there is only one plausible measure (non-mutual recursion with
only one varying
parameter), just use that measure.
Side benefit: If the function isn’t terminating, more detailed error
messages are shown
(failing proof goals), located at the recursive calls.
This improves Lean’s capabilities to guess the termination measure for
well-founded
recursion, by also trying lexicographic orders. For example:
def ackermann (n m : Nat) := match n, m with
| 0, m => m + 1
| .succ n, 0 => ackermann n 1
| .succ n, .succ m => ackermann n (ackermann (n + 1) m)
now just works.
The module docstring of `Lean.Elab.PreDefinition.WF.GuessLex` tells the
technical story.
Fixes#2837
