Piotr Osiewicz
d338e4a8a6
# Background
In https://github.com/zed-industries/zed/issues/14408 we received a
repro for "Replace all" being slow, even after the work I did
https://github.com/zed-industries/zed/pull/13654. Admittedly #13654 was
a pretty straightforward change.
Under the profiler it turned out that we're spending *10 seconds* in
`memmove` on main thread. Ugh. Not great. The direct ancestor of the
memmove call was
https://github.com/zed-industries/zed/blob/66f0c390a88f5aa16d21047a8982a5888fa0f816/crates/editor/src/display_map/tab_map.rs#L108-L119
What?
# Accidental O(n^2)
We have a bunch of `consolidate_*_edits` functions which take a list of
Fold/Tab/Inlay/Wrap edits and merge consecutive edits if their ranges
overlap/are next to one another. The loop usually goes as follows:
```
while ix < edits.len() {
let (prev_edits, next_edits) = edits.split_at_mut(ix);
let prev_edit = prev_edits.last_mut().unwrap();
let edit = &next_edits[0];
if PREV_EDIT_CAN_BE_MERGED_WITH_CURRENT_ONE {
MERGE_EDITS(prev_edit, edit);
edits.remove(ix); // !!
} else {
ix += 1;
}
}
```
The problem is the call to `.remove` - it has to shift all of the
consecutive elements in the `edits` vector! Thus, when processing the
edits from the original repro (where consolidation shrinks the edit list
from 210k entries to 30k), we mostly spend time moving entries in memory
around.
Thus, the original repro isn't really an issue with replace_all; it's
just that replace_all is one of the few tools available to the end user
that can apply large # of edits in a single transaction.
# Solution
In this PR I address the issue by rewriting the loop in a way that does
not throw items away via `.remove`. Instead, `Iterator::scan` is used,
which lets us achieve the same logic without having the pitfalls of
`.remove`s.
Crucially, **this code does not allocate a new backing buffer for
edits** (see [this article for
rationale](https://blog.polybdenum.com/2024/01/17/identifying-the-collect-vec-memory-leak-footgun.html));
with `vec.into_iter().scan().filter_map().collect()` we still use the
same underlying buffer as the one that's passed into `consolidate_*`
functions. In development I verified that by checking whether the
pointers to backing storage of a Vec are the same before and after the
consolidation.
# Results
### Before
Nightly 0.145.0
[66f0c390a8](https://github.com/zed-industries/zed/commit/66f0c390a88f5aa16d21047a8982a5888fa0f816)
https://github.com/user-attachments/assets/8b0ad3bc-86d6-4f8a-850c-ebb86e8b3bfc
(~13s end-to-end)
### After
https://github.com/user-attachments/assets/366835db-1d84-4f95-8c74-b1506a9fabec
(~2s end-to-end)
The remaining lag is (I think) lies in `TextSummary` calculation and not
the consolidation itself. Thus, for the purposes of scoping this PR,
I'll tackle it separately.
Release Notes:
- Significantly improved performance of applying large quantities of
concurrent edits (e.g. when running "Replace all").
Zed
Welcome to Zed, a high-performance, multiplayer code editor from the creators of Atom and Tree-sitter.
Installation
On macOS and Linux you can download Zed directly or install Zed via your local package manager.
Other platforms are not yet available:
- Windows (tracking issue)
- Web (tracking issue)
Developing Zed
- Building Zed for macOS
- Building Zed for Linux
- Building Zed for Windows
- Running Collaboration Locally
Contributing
See CONTRIBUTING.md for ways you can contribute to Zed.
Also... we're hiring! Check out our jobs page for open roles.
Licensing
License information for third party dependencies must be correctly provided for CI to pass.
We use cargo-about to automatically comply with open source licenses. If CI is failing, check the following:
- Is it showing a
no license specifiederror for a crate you've created? If so, addpublish = falseunder[package]in your crate's Cargo.toml. - Is the error
failed to satisfy license requirementsfor a dependency? If so, first determine what license the project has and whether this system is sufficient to comply with this license's requirements. If you're unsure, ask a lawyer. Once you've verified that this system is acceptable add the license's SPDX identifier to theacceptedarray inscript/licenses/zed-licenses.toml. - Is
cargo-aboutunable to find the license for a dependency? If so, add a clarification field at the end ofscript/licenses/zed-licenses.toml, as specified in the cargo-about book.