Reduce memory consumption by sharing the previous dependency graph's
edges with the current graph when it is known to be valid to do so. It
is known to be valid whenever we mark a node green because all of its
dependencies were green. It is *not* known to be valid when we mark a
node green because we re-executed its query and its result was the same
as in the previous compilation session. In that case, the dependency set
might have changed (we don't try to determine whether or not it changed
and whether or not we can share).
Reduce memory consumption by taking advantage of red/green algorithm
properties to share the previous dependency graph's node data with the
current graph instead of storing node data redundantly. Red nodes can
share the `DepNode`, and green nodes can share the `DepNode` and
`Fingerprint`. Edges will be shared when possible in a later change.
Clarify constructor splitting in exhaustiveness checking
I reworked the explanation of the algorithm completely to make it properly account for the various extensions we've added. This includes constructor splitting, which was previously not clearly included in the algorithm. This makes wildcards less magical; I added some detailed examples; and this distinguishes clearly between constructors that only make sense in patterns (like ranges) and those that make sense for values (like `Some`). This reformulation had been floating around in my mind for a while, and I'm quite happy with how it turned out. Let me know how you feel about it.
I also factored out all three cases of splitting (wildcards, ranges and slices) into dedicated structs to encapsulate the complicated bits.
I measured no perf impact but I don't trust my local measurements for refactors since https://github.com/rust-lang/rust/pull/79284.
r? `@varkor`
`@rustbot` modify labels: +A-exhaustiveness-checking
rustc_query_system: explicitly register reused dep nodes
Register nodes that we've reused from the previous session explicitly
with `OnDiskCache`. Previously, we relied on this happening as a side
effect of accessing the nodes in the `PreviousDepGraph`. For the sake of
performance and avoiding unintended side effects, register explictily.
This implements support for applying PGO to the rustc compilation step (not
standard library or any tooling, including rustdoc). Expanding PGO to more tools
is not terribly difficult but will involve more work and greater CI time
commitment.
For the same reason of avoiding greater time commitment, this currently avoids
implementing for platforms outside of x86_64-unknown-linux-gnu, though in
practice it should be quite simple to extend over time to more platforms. The
initial implementation is intentionally minimal here to avoid too much work
investment before we start seeing wins for a subset of Rust users.
The choice of workloads to profile here is somewhat arbitrary, but the general
rationale was to aim for a small set that largely avoided time regressions on
perf.rust-lang.org's full suite of crates. The set chosen is libcore, cargo (and
its dependencies), and a few ad-hoc stress tests from perf.rlo. The stress tests
are arguably the most controversial, but they benefit those cases (avoiding
regressions) and do not really remove wins from other benchmarks.
The primary next step after this PR lands is to implement support for PGO in
LLVM. It is unclear whether we can afford a full LLVM rebuild in CI, though, so
the approach taken there may need to be more staggered. rustc-only PGO seems
well affordable on linux at least, giving us up to 20% wall time wins on some
crates for 15 minutes of extra CI time (1 hour up from 45 minutes).
The PGO data is uploaded to allow others to reuse it if attempting to reproduce
the CI build or potentially, in the future, on other platforms where an
off-by-one strategy is used for dist builds at minimal performance cost.
This caught several bugs where people expected `slice` to link to the
primitive, but it linked to the module instead.
This also uses `cfg_attr(bootstrap)` since the ambiguity only occurs
when compiling with stage 1.
This does *not* currently work for associated items that are
auto-implemented by the compiler (e.g. `never::eq`), because they aren't
present in the source code. I plan to fix this in a follow-up PR.
`replace_prefix` is currently implemented as a method but has no real relation
to the struct it is implemented on. Turn it into a closure and move it into the
only method from which it is called.
Added [T; N]::zip()
This is my first PR to rust so I hope I have done everything right, or at least close :)
---
This is PR adds the array method `[T; N]::zip()` which, in my mind, is a natural extension to #75212.
My implementation of `zip()` is mostly just a modified copy-paste of `map()`. Should I keep the comments? Also am I right in assuming there should be no way for the `for`-loop to panic, thus no need for the dropguard seen in the `map()`-function?
The doc comment is in a similar way a slightly modified copy paste of [`Iterator::zip()`](https://doc.rust-lang.org/beta/std/iter/trait.Iterator.html#method.zip)
`@jplatte` mentioned in [#75490](https://github.com/rust-lang/rust/pull/75490#issuecomment-677790758) `zip_with()`,
> zip and zip_with seem like they would be useful :)
is this something I should add (assuming there is interest for this PR at all :))
Rename `overlapping_patterns` lint
As discussed in https://github.com/rust-lang/rust/issues/65477. I also tweaked a few things along the way.
r? `@varkor`
`@rustbot` modify labels: +A-exhaustiveness-checking
Reserve necessary space for params in generics_of
Always reserve space for the exact number of generic parameters we need in generics_of. As far as I can see, the default is 0/4 elements based on has_self, and the vector grows on after that.
