rustdoc search: prefer stable items in search results
fixes https://github.com/rust-lang/rust/issues/138067
this does add a new field to the search index, but since we're only listing unstable items instead of adding a boolean flag to every item, it should only increase the search index size of sysroot crates, since those are the only ones using the `staged_api` feature, at least as far as the rust project is concerned.
`concat_idents!` is in the process of being removed, but a few things it
is used to test will still be relevant. Migrate these tests to something
other than `concat_idents`.
rustdoc-search: handle `impl Into<X>` better
This PR fixes two bugs I ran into while searching the compiler docs:
- It omitted an `impl Trait` entry in the type signature field, producing `TyCtxt, , Symbol -> bool`
- It didn't let me search for `TyCtxt, DefId, Symbol -> bool` even though that's a perfectly good description of the function I was looking for (the function actually used `impl Into<DefId>`
r? ``@GuillaumeGomez`` cc ``@lolbinarycat``
While normal generics can be skipped in this case, no-names need
something to show here.
Before: `TyCtxt, , Symbol -> bool`
After: `TyCtxt, Into<DefId>, Symbol -> bool`
This change uses the same "exact" paths as trait implementors
and type alias inlining to track items with multiple
reachable paths. This way, if you search for `vec`, you get
only the `std` exports of it, and not the one from `alloc`.
It still includes all the items in the search index so that
you can search for them by all available paths. For example,
try `core::option` and `std::option`, and notice that the
results page doesn't show duplicates, but still shows all
the items in their respective crates.
rustdoc-search: depth limit `T<U>` -> `U` unboxing
Profiler output:
https://notriddle.com/rustdoc-html-demo-9/search-unbox-limit/ (the only significant change is that one of the `rust` tests went from 378416ms to 16ms).
This is a performance enhancement aimed at a problem I found while using type-driven search on the Rust compiler. It is caused by [`Interner`], a trait with 41 associated types, many of which recurse back to `Self` again.
This caused search.js to struggle. It eventually terminates, after about 10 minutes of turning my PC into a space header, but it's doing `41!` unifications and that's too slow.
[`Interner`]: https://doc.rust-lang.org/nightly/nightly-rustc/rustc_middle/ty/trait.Interner.html
This is implemented, in addition to the ML-style one,
because Rust does it. If we don't, we'll never hear the end of it.
This commit also refactors some duplicate parts of the parser
into a dedicated function.
Option::map, for example, looks like this:
option<t>, (t -> u) -> option<u>
This syntax searches all of the HOFs in Rust: traits Fn, FnOnce,
and FnMut, and bare fn primitives.
rustdoc: search for tuples and unit by type with `()`
This feature extends rustdoc to support the syntax that most users will naturally attempt to use to search for tuples. Part of https://github.com/rust-lang/rust/issues/60485
Function signature searches already support tuples and unit. The explicit name `primitive:tuple` and `primitive:unit` can be used to match a tuple or unit, while `()` will match either one. It also follows the direction set by the actual language for parens as a group, so `(u8,)` will only match a tuple, while `(u8)` will match a plain, unwrapped byte—thanks to loose search semantics, it will also match the tuple.
## Preview
* [`option<t>, option<u> -> (t, u)`](<https://notriddle.com/rustdoc-html-demo-5/tuple-unit/std/index.html?search=option%3Ct%3E%2C option%3Cu%3E -%3E (t%2C u)>)
* [`[t] -> (t,)`](<https://notriddle.com/rustdoc-html-demo-5/tuple-unit/std/index.html?search=[t] -%3E (t%2C)>)
* [`(ipaddr,) -> socketaddr`](<https://notriddle.com/rustdoc-html-demo-5/tuple-unit/std/index.html?search=(ipaddr%2C) -%3E socketaddr>)
## Motivation
When type-based search was first landed, it was directly [described as incomplete][a comment].
[a comment]: https://github.com/rust-lang/rust/pull/23289#issuecomment-79437386
Filling out the missing functionality is going to mean adding support for more of Rust's [type expression] syntax, such as tuples (in this PR), references, raw pointers, function pointers, and closures.
[type expression]: https://doc.rust-lang.org/reference/types.html#type-expressions
There does seem to be demand for this sort of thing, such as [this Discord message](https://discord.com/channels/442252698964721669/443150878111694848/1042145740065099796) expressing regret at rustdoc not supporting tuples in search queries.
## Reference description (from the Rustdoc book)
<table>
<thead>
<tr>
<th>Shorthand</th>
<th>Explicit names</th>
</tr>
</thead>
<tbody>
<tr><td colspan="2">Before this PR</td></tr>
<tr>
<td><code>[]</code></td>
<td><code>primitive:slice</code> and/or <code>primitive:array</code></td>
</tr>
<tr>
<td><code>[T]</code></td>
<td><code>primitive:slice<T></code> and/or <code>primitive:array<T></code></td>
</tr>
<tr>
<td><code>!</code></td>
<td><code>primitive:never</code></td>
</tr>
<tr><td colspan="2">After this PR</td></tr>
<tr>
<td><code>()</code></td>
<td><code>primitive:unit</code> and/or <code>primitive:tuple</code></td>
</tr>
<tr>
<td><code>(T)</code></td>
<td><code>T</code></td>
</tr>
<tr>
<td><code>(T,)</code></td>
<td><code>primitive:tuple<T></code></td>
</tr>
</tbody>
</table>
A single type expression wrapped in parens is the same as that type expression, since parens act as the grouping operator. If they're empty, though, they will match both `unit` and `tuple`, and if there's more than one type (or a trailing or leading comma) it is the same as `primitive:tuple<...>`.
However, since items can be left out of the query, `(T)` will still return results for types that match tuples, even though it also matches the type on its own. That is, `(u32)` matches `(u32,)` for the exact same reason that it also matches `Result<u32, Error>`.
## Future direction
The [type expression grammar](https://doc.rust-lang.org/reference/types.html#type-expressions) from the Reference is given below:
<pre><code>Syntax
Type :
TypeNoBounds
| <a href="https://doc.rust-lang.org/reference/types/impl-trait.html">ImplTraitType</a>
| <a href="https://doc.rust-lang.org/reference/types/trait-object.html">TraitObjectType</a>
<br>
TypeNoBounds :
<a href="https://doc.rust-lang.org/reference/types.html#parenthesized-types">ParenthesizedType</a>
| <a href="https://doc.rust-lang.org/reference/types/impl-trait.html">ImplTraitTypeOneBound</a>
| <a href="https://doc.rust-lang.org/reference/types/trait-object.html">TraitObjectTypeOneBound</a>
| <a href="https://doc.rust-lang.org/reference/paths.html#paths-in-types">TypePath</a>
| <a href="https://doc.rust-lang.org/reference/types/tuple.html#tuple-types">TupleType</a>
| <a href="https://doc.rust-lang.org/reference/types/never.html">NeverType</a>
| <a href="https://doc.rust-lang.org/reference/types/pointer.html#raw-pointers-const-and-mut">RawPointerType</a>
| <a href="https://doc.rust-lang.org/reference/types/pointer.html#shared-references-">ReferenceType</a>
| <a href="https://doc.rust-lang.org/reference/types/array.html">ArrayType</a>
| <a href="https://doc.rust-lang.org/reference/types/slice.html">SliceType</a>
| <a href="https://doc.rust-lang.org/reference/types/inferred.html">InferredType</a>
| <a href="https://doc.rust-lang.org/reference/paths.html#qualified-paths">QualifiedPathInType</a>
| <a href="https://doc.rust-lang.org/reference/types/function-pointer.html">BareFunctionType</a>
| <a href="https://doc.rust-lang.org/reference/macros.html#macro-invocation">MacroInvocation</a>
</code></pre>
ImplTraitType and TraitObjectType (and ImplTraitTypeOneBound and TraitObjectTypeOneBound) are not yet implemented. They would mostly desugar to `trait:`, similarly to how `!` desugars to `primitive:never`.
ParenthesizedType and TuplePath are added in this PR.
TypePath is already implemented (except const generics, which is not planned, and function-like trait syntax, which is planned as part of closure support).
NeverType is already implemented.
RawPointerType and ReferenceType require parsing and fixes to the search index to store this information, but otherwise their behavior seems simple enough. Just like tuples and slices, `&T` would be equivalent to `primitive:reference<T>`, `&mut T` would be equivalent to `primitive:reference<keyword:mut, T>`, `*T` would be equivalent to `primitive:pointer<T>`, `*mut T` would be equivalent to `primitive:pointer<keyword:mut, T>`, and `*const T` would be equivalent to `primitive:pointer<keyword:const, T>`. Lifetime generics support is not planned, because lifetime subtyping seems too complicated.
ArrayType is subsumed by SliceType right now. Implementing const generics is not planned, because it seems like it would require a lot of implementation complexity for not much gain.
InferredType isn't really covered right now. Its semantics in a search context are not obvious.
QualifiedPathInType is not implemented, and it is not planned. I would need a use case to justify it, and act as a guide for what the exact semantics should be.
BareFunctionType is not implemented. Along with function-like trait syntax, which is formally considered a TypePath, it's the biggest missing feature to be able to do structured searches over generic APIs like `Option`.
MacroInvocation is not parsed (macro names are, but they don't mean the same thing here at all). Those are gone by the time Rustdoc sees the source code.
Since the two are counted separately elsewhere, they should get
their own limits, too. The biggest problem with combining them
is that paths are loosely checked by not requiring every component
to match, which means that if they are short and matched loosely,
they can easily find "drunk typist" matches that make no sense,
like this old result:
std::collections::btree_map::itermut matching slice::itermut
maxEditDistance = ("slice::itermut".length) / 3 = 14 / 3 = 4
editDistance("std", "slice") = 4
editDistance("itermut", "itermut") = 0
4 + 0 <= 4 PASS
Of course, `slice::itermut` should not match stuff from btreemap.
`slice` should not match `std`.
The new result counts them separately:
maxPathEditDistance = "slice".length / 3 = 5 / 3 = 1
maxEditDistance = "itermut".length / 3 = 7 / 3 = 2
editDistance("std", "slice") = 4
4 <= 1 FAIL
Effectively, this makes path queries less "typo-resistant".
It's not zero, but it means `vec` won't match the `v1` prelude.
Queries without parent paths are unchanged.
This might have made sense if the algorithm could use `searchWords`
to skip having to look at `searchIndex`, but since it always
does a substring check on both the stock word and the normalizedName,
it doesn't seem to help performance anyway.
This commit adds ranking and quick filtering to type-based search,
improving performance and having it order results based on their
type signatures.
Motivation
----------
If I write a query like `str -> String`, a lot of functions come up.
That's to be expected, but `String::from_str` should come up on top, and
it doesn't right now. This is because the sorting algorithm is based
on the functions name, and doesn't consider the type signature at all.
`slice::join` even comes up above it!
To fix this, the sorting should take into account the function's
signature, and the closer match should come up on top.
Guide-level description
-----------------------
When searching by type signature, types with a "closer" match will
show up above types that match less precisely.
Reference-level explanation
---------------------------
Functions signature search works in three major phases:
* A compact "fingerprint," based on the [bloom filter] technique, is used to
check for matches and to estimate the distance. It sometimes has false
positive matches, but it also operates on 128 bit contiguous memory and
requires no backtracking, so it performs a lot better than real
unification.
The fingerprint represents the set of items in the type signature, but it
does not represent nesting, and it ignores when the same item appears more
than once.
The result is rejected if any query bits are absent in the function, or
if the distance is higher than the current maximum and 200
results have already been found.
* The second step performs unification. This is where nesting and true bag
semantics are taken into account, and it has no false positives. It uses a
recursive, backtracking algorithm.
The result is rejected if any query elements are absent in the function.
[bloom filter]: https://en.wikipedia.org/wiki/Bloom_filter
Drawbacks
---------
This makes the code bigger.
More than that, this design is a subtle trade-off. It makes the cases I've
tested against measurably faster, but it's not clear how well this extends
to other crates with potentially more functions and fewer types.
The more complex things get, the more important it is to gather a good set
of data to test with (this is arguably more important than the actual
benchmarking ifrastructure right now).
Rationale and alternatives
--------------------------
Throwing a bloom filter in front makes it faster.
More than that, it tries to take a tactic where the system can not only check
for potential matches, but also gets an accurate distance function without
needing to do unification. That way it can skip unification even on items
that have the needed elems, as long as they have more items than the
currently found maximum.
If I didn't want to be able to cheaply do set operations on the fingerprint,
a [cuckoo filter] is supposed to have better performance.
But the nice bit-banging set intersection doesn't work AFAIK.
I also looked into [minhashing], but since it's actually an unbiased
estimate of the similarity coefficient, I'm not sure how it could be used
to skip unification (I wouldn't know if the estimate was too low or
too high).
This function actually uses the number of distinct items as its
"distance function."
This should give the same results that it would have gotten from a Jaccard
Distance $1-\frac{|F\cap{}Q|}{|F\cup{}Q|}$, while being cheaper to compute.
This is because:
* The function $F$ must be a superset of the query $Q$, so their union is
just $F$ and the intersection is $Q$ and it can be reduced to
$1-\frac{|Q|}{|F|}.
* There are no magic thresholds. These values are only being used to
compare against each other while sorting (and, if 200 results are found,
to compare with the maximum match). This means we only care if one value
is bigger than the other, not what it's actual value is, and since $Q$ is
the same for everything, it can be safely left out, reducing the formula
to $1-\frac{1}{|F|} = \frac{|F|}{|F|}-\frac{1}{|F|} = |F|-1$. And, since
the values are only being compared with each other, $|F|$ is fine.
Prior art
---------
This is significantly different from how Hoogle does it.
It doesn't account for order, and it has no special account for nesting,
though `Box<t>` is still two items, while `t` is only one.
This should give the same results that it would have gotten from a Jaccard
Distance $1-\frac{|A\cap{}B|}{|A\cup{}B|}$, while being cheaper to compute.
Unresolved questions
--------------------
`[]` and `()`, the slice/array and tuple/union operators, are ignored while
building the signature for the query. This is because they match more than
one thing, making them ambiguous. Unfortunately, this also makes them
a performance cliff. Is this likely to be a problem?
Right now, the system just stashes the type distance into the
same field that levenshtein distance normally goes in. This means exact
query matches show up on top (for example, if you have a function like
`fn nothing(a: Nothing, b: i32)`, then searching for `nothing` will show it
on top even if there's another function with `fn bar(x: Nothing)` that's
technically a closer match in type signature.
Future possibilities
--------------------
It should be possible to adopt more sorting criteria to act as a tie breaker,
which could be determined during unification.
[cuckoo filter]: https://en.wikipedia.org/wiki/Cuckoo_filter
[minhashing]: https://en.wikipedia.org/wiki/MinHash
