rustdoc: Add unstable option to only emit shared/crate-specific files
The intended use case is for docs.rs, which can now copy exactly the
files it cares about, rather than having to guess based on whether they
have a resource suffix or not. In particular, some files have a resource
suffix but cannot be shared between crates: https://github.com/rust-lang/docs.rs/pull/1312#issuecomment-798783688
The end goal is to fixrust-lang/docs.rs#1327 by reverting rust-lang/docs.rs#1324.
This obsoletes `--print=unversioned-files`, which I plan to remove as
soon as docs.rs stops using it.
I recommend reviewing this one commit at a time.
r? ``@GuillaumeGomez`` cc ``@Nemo157`` ``@pietroalbini``
Remove unnecessary `Option` wrapping around `Crate.module`
I'm wondering if it was originally there so that we could `take` the
module which enables `after_krate` to take an `&Crate`. However, the two
impls of `after_krate` only use `Crate.name`, so we can pass just the
name instead.
Sidebar trait items order
We were actually sorting `Symbol` and not `String`, creating a completely invalid sort result. I added a test to prevent regressions.
r? ``@jyn514``
Rename `source` to `span` and `span` to `source`
- Rename `clean::Item.source` to `span`
- Rename `clean::Span::span()` to `clean::Span::inner()`
- Rename `rustdoc_json_types::Item.source` to `span`
- rustdoc-json: Rename `Import.span` to `Import.source`
*See also the [discussion on Zulip][z] (this is a bit more than discussed in
that conversation, but all the changes are related).*
r? `@jyn514`
[z]: https://rust-lang.zulipchat.com/#narrow/stream/182449-t-compiler.2Fhelp/topic/get.20span.20of.20file.20from.20name/near/229603729
I'm wondering if it was originally there so that we could `take` the
module which enables `after_krate` to take an `&Crate`. However, the two
impls of `after_krate` only use `Crate.name`, so we can pass just the
name instead.
Remove theme.js file
Fixes#82616.
The first commit moves the `theme.js` file into `main.js`, which requires to also run a small `.replace` on the `main.js` content.
The second commit is just a small cleanup to centralize DOM ids.
Since it removes a file from rustdoc output: cc `@rust-lang/docs-rs`
cc `@jsha`
r? `@jyn514`
Avoid temporary allocations in `render_assoc_item`
`render_assoc_item` came up as very hot in a profile of rustdoc on
`bevy`. This avoids some temporary allocations just to calculate the
length of the header.
This should be a strict improvement, since all string formatting was
done twice before.
cc #82845
Otherwise you get a lot of instances of `item.span.span()`, which is
just plain confusing. `item.span.inner()` conveys the correct meaning of
"get the type that `clean::Span` wraps".
This essentially switches search-index.js from a "array of struct"
to a "struct of array" format, like this:
{
"doc": "Crate documentation",
"t": [ 1, 1, 2, 3, ... ],
"n": [ "Something", "SomethingElse", "whatever", "do_stuff", ... ],
"q": [ "a::b", "", "", "", ... ],
"d": [ "A Struct That Does Something", "Another Struct", "a function", "another function", ... ],
"i": [ 0, 0, 1, 1, ... ],
"f": [ null, null, [], [], ... ],
"p": ...,
"a": ...
}
So `{ty: 1, name: "Something", path: "a::b", desc: "A Struct That Does Something", parent_idx: 0, search_type: null}` is the first item.
This makes the uncompressed version smaller, but it really shows on the
compressed version:
notriddle:rust$ wc -c new-search-index1.52.0.js
2622427 new-search-index1.52.0.js
notriddle:rust$ wc -c old-search-index1.52.0.js
2725046 old-search-index1.52.0.js
notriddle:rust$ gzip new-search-index1.52.0.js
notriddle:rust$ gzip old-search-index1.52.0.js
notriddle:rust$ wc -c new-search-index1.52.0.js.gz
239385 new-search-index1.52.0.js.gz
notriddle:rust$ wc -c old-search-index1.52.0.js.gz
296328 old-search-index1.52.0.js.gz
notriddle:rust$
That's a 4% improvement on the uncompressed version (fewer `[]`),
and 20% improvement after gzipping it, thanks to putting like-typed
data next to each other. Any compression algorithm based on a sliding
window will probably show this kind of improvement.
rustdoc: Cleanup `html::render::Context`
- Move most shared fields to `SharedContext` (except for `cache`, which
isn't mutated anyway)
- Replace a use of `Arc` with `Rc`
- Make a bunch of fields private
- Add static size assertion for `Context`
- Don't share `id_map` and `deref_id_map`
`render_assoc_item` came up as very hot in a profile of rustdoc on
`bevy`. This avoids some temporary allocations just to calculate the
length of the header.
This should be a strict improvement, since all string formatting was
done twice before.
rustdoc: Add an unstable option to print all unversioned files
This allows sharing those files between different doc invocations
without having to know their names ahead of time.
Helps with https://github.com/rust-lang/docs.rs/issues/1302.
r? ````@GuillaumeGomez```` cc ````@pietroalbini```` ````@Nemo157````
I don't think the boxing helped performance, in fact I think it
potentially made it worse. The data was still being copied, but now it
was through a pointer. Thinking about it more, I think boxing might only
help when you're passing a big object around by value all the time,
rather than the slowdown being that you're cloning it.
There was no need to clone `id_map` because it was reset before each
item was rendered. `deref_id_map` was not reset, but it was keyed by
`DefId` and thus was unlikely to have collisions (at least for now).
Now we just clone the fields that need to be cloned, and instead create
fresh versions of the others.
All the tests passed, so it doesn't seem they need to be shared.
Plus they should be item/page-specific.
I'm not sure why they were shared before. I think the reason `id_map`
worked as a shared value before is that it is cleared before rendering
each item (in `render_item`). And then I'm guessing `deref_id_map`
worked because it's a hashmap keyed by `DefId`, so there was no overlap
(though I'm guessing we could have had issues in the future).
Note that `id_map` currently still has to be cleared because otherwise
child items would inherit the `id_map` of their parent. I'm hoping to
figure out a way to stop cloning `Context`, but until then we have to
reset `id_map`.
It doesn't look like it's shared across threads, so it doesn't need to
be thread-safe. Of course, since we're using Rust, we'll get an error if
we try to share it across threads, so this should be safe :)
Instead of being loaded on every page, the JS search index is now
loaded when either (a) there is a `?search=` param, or (b) the search
input is focused.
This saves both CPU and bandwidth. As of Feb 2021,
https://doc.rust-lang.org/search-index1.50.0.js is 273,838 bytes
gzipped or 2,544,939 bytes uncompressed. Evaluating it takes 445 ms
of CPU time in Chrome 88 on a i7-10710U CPU (out of a total ~2,100
ms page reload).
Generate separate JS file with crate names.
This is much smaller than the full search index, and is used in the "hot
path" to draw the page. In particular it's used to crate the dropdown
for the search bar, and to append a list of crates to the sidebar (on
some pages).
Skip early search that can bypass 500ms timeout.
This was occurring when someone had typed some text during the load of
search-index.js. Their query was usually not ready to execute, and the
search itself is fairly expensive, delaying the overall load, which
delayed the input / keyup events, which delayed eventually executing the
query.
