Use escape_default() for strings in LitKind::token(). #50391

nnethercote · 2018-05-02T10:43:43Z

This avoids converting every char to \u{...} form, which bloats the
resulting strings unnecessarily. It also provides consistency with the
existing escape_default() calls in LitKind::token() used for raw
string literals, char literals, and raw byte char literals.

There are two benefits from this change.

Compilation is faster. Most of the rustc-perf benchmarks see a
non-trivial speedup, particularly for incremental rebuilds, with the
best speedup over 13%, and multiple others over 10%.
Generated rlibs are smaller. An extreme example is libfutures.rlib,
which shrinks from 2073306 bytes to 1765927 bytes, a 15% reduction.

r? @jseyfried

Here are full numbers for all the rustc-perf runs where the improvement was > 1%.

regex-check
	avg: -11.1%	min: -13.4%	max: -5.5%
futures-check
	avg: -7.6%	min: -11.4%	max: -3.5%
futures-opt
	avg: -6.3%	min: -10.3%	max: -2.3%
futures
	avg: -6.6%	min: -10.3%	max: -2.8%
regex-opt
	avg: -4.7%	min: -10.2%	max: -0.4%
regex
	avg: -5.3%	min: -10.2%	max: -1.2%
hyper-check
	avg: -4.8%	min: -6.6%	max: -2.7%
encoding-check
	avg: -4.1%	min: -5.5%	max: -2.5%
issue-46449-check
	avg: -4.7%	min: -5.2%	max: -4.1%
clap-rs-check
	avg: -2.9%	min: -5.2%	max: -1.1%
hyper
	avg: -3.0%	min: -5.1%	max: -0.8%
parser-check
	avg: -4.2%	min: -4.9%	max: -3.2%
hyper-opt
	avg: -2.6%	min: -4.9%	max: -0.3%
encoding-opt
	avg: -2.3%	min: -4.6%	max: -0.5%
encoding
	avg: -2.5%	min: -4.4%	max: -0.6%
issue-46449
	avg: -2.3%	min: -4.4%	max: -1.8%
issue-46449-opt
	avg: -1.7%	min: -4.3%	max: -0.9%
clap-rs-opt
	avg: -1.6%	min: -4.2%	max: -0.2%
serde-check
	avg: -1.4%	min: -4.1%	max: -0.2%
clap-rs
	avg: -1.6%	min: -3.9%	max: -0.7%
unify-linearly-check
	avg: -3.2%	min: -3.7%	max: -2.7%
serde
	avg: -1.1%	min: -3.5%	max: -0.1%
regression-31157-check
	avg: -2.6%	min: -3.4%	max: -1.6%
helloworld-check
	avg: -2.5%	min: -3.4%	max: -0.6%
serde-opt
	avg: -1.3%	min: -3.3%	max: -0.5%
tokio-webpush-simple-check
	avg: -2.4%	min: -3.2%	max: -1.8%
piston-image-check
	avg: -1.7%	min: -3.2%	max: -0.9%
deeply-nested-opt
	avg: -1.5%	min: -3.0%	max: -0.6%
deeply-nested-check
	avg: -1.9%	min: -2.9%	max: -0.4%
deeply-nested
	avg: -1.9%	min: -2.9%	max: -1.2%
syn-check
	avg: -1.8%	min: -2.8%	max: -0.6%
coercions
	avg: -0.5%	min: -2.8%	max: 0.4%
syn-opt
	avg: -0.9%	min: -2.4%	max: -0.1%
syn
	avg: -1.1%	min: -2.2%	max: -0.3%
parser-opt
	avg: -1.9%	min: -2.1%	max: -1.6%
parser
	avg: -1.9%	min: -2.1%	max: -1.6%
style-servo-check
	avg: -1.3%	min: -2.0%	max: -0.8%
regression-31157-opt
	avg: -0.8%	min: -2.0%	max: 0.0%
piston-image
	avg: -0.7%	min: -1.8%	max: -0.2%
piston-image-opt
	avg: -0.6%	min: -1.8%	max: -0.0%
regression-31157
	avg: -1.0%	min: -1.7%	max: -0.3%
html5ever-opt
	avg: -0.6%	min: -1.5%	max: -0.1%
unify-linearly-opt
	avg: -1.3%	min: -1.5%	max: -1.1%
unify-linearly
	avg: -1.3%	min: -1.4%	max: -1.2%
tokio-webpush-simple-opt
	avg: -0.4%	min: -1.2%	max: -0.0%
helloworld-opt
	avg: -1.0%	min: -1.1%	max: -0.6%
helloworld
	avg: -1.0%	min: -1.1%	max: -0.7%
inflate-opt
	avg: -0.3%	min: -1.1%	max: 0.1%
html5ever-check
	avg: -0.6%	min: -1.0%	max: -0.3%
inflate-check
	avg: -0.3%	min: -1.0%	max: -0.1%

kennytm · 2018-05-02T12:21:09Z

src/libsyntax/attr.rs

@@ -1230,7 +1230,7 @@ impl LitKind {
            LitKind::Str(string, ast::StrStyle::Cooked) => {
                let mut escaped = String::new();
                for ch in string.as_str().chars() {
-                    escaped.extend(ch.escape_unicode());
+                    escaped.extend(ch.escape_default());
                }


We could use str::escape_default nowadays.

let escaped = string.escape_default();

(There is also a fn escape_default(s: &str) -> String in src/libsyntax/parse.rs but that should probably be removed in favor of the std one.)

However, I think only \r, \, and " really needs to be escaped. All other characters can legally appear inside a string. This may provide even more speed up since we don't need the UTF-8 encode/decode step.

Note that this function is not the hot one. str_lit(), which parses the string produced here, is the hot function. So str::escape_default() might be more concise, but it's unlikely to make things faster.

Also, in practice these strings are all rustdoc comments, and escaped chars are rare. So it also doesn't matter much exactly which characters we escape.

eddyb · 2018-05-02T14:08:49Z

r=me with @kennytm's suggestion

nnethercote · 2018-05-02T23:52:57Z

@eddyb: I switched to str::escape_default. I also added a second commit that removes parse::escape_default.

kennytm · 2018-05-02T23:59:37Z

src/libsyntax/attr.rs

-                for ch in string.as_str().chars() {
-                    escaped.extend(ch.escape_unicode());
-                }
+                let mut escaped = string.as_str().escape_default();


mut is no longer needed.

This avoids converting every char to \u{...} form, which bloats the resulting strings unnecessarily. It also provides consistency with the existing escape_default() calls in LitKind::token() used for raw string literals, char literals, and raw byte char literals. There are two benefits from this change. - Compilation is faster. Most of the rustc-perf benchmarks see a non-trivial speedup, particularly for incremental rebuilds, with the best speedup over 13%, and multiple others over 10%. - Generated rlibs are smaller. An extreme example is libfutures.rlib, which shrinks from 2073306 bytes to 1765927 bytes, a 15% reduction.

str::escape_default() can be used instead.

nnethercote · 2018-05-03T01:12:29Z

I removed the unnecessary mut. (I'm surprised rustc didn't complain about it...)

kennytm · 2018-05-03T01:13:46Z

@bors r=eddyb

bors · 2018-05-03T01:13:47Z

📌 Commit 7a56360 has been approved by eddyb

bors · 2018-05-03T08:06:16Z

⌛ Testing commit 7a56360 with merge 698b956...

@jseyfried

Use escape_default() for strings in LitKind::token(). This avoids converting every char to \u{...} form, which bloats the resulting strings unnecessarily. It also provides consistency with the existing escape_default() calls in LitKind::token() used for raw string literals, char literals, and raw byte char literals. There are two benefits from this change. - Compilation is faster. Most of the rustc-perf benchmarks see a non-trivial speedup, particularly for incremental rebuilds, with the best speedup over 13%, and multiple others over 10%. - Generated rlibs are smaller. An extreme example is libfutures.rlib, which shrinks from 2073306 bytes to 1765927 bytes, a 15% reduction. r? @jseyfried <details><summary>Here are full numbers for all the rustc-perf runs where the improvement was > 1%.</summary> ``` regex-check avg: -11.1% min: -13.4% max: -5.5% futures-check avg: -7.6% min: -11.4% max: -3.5% futures-opt avg: -6.3% min: -10.3% max: -2.3% futures avg: -6.6% min: -10.3% max: -2.8% regex-opt avg: -4.7% min: -10.2% max: -0.4% regex avg: -5.3% min: -10.2% max: -1.2% hyper-check avg: -4.8% min: -6.6% max: -2.7% encoding-check avg: -4.1% min: -5.5% max: -2.5% issue-46449-check avg: -4.7% min: -5.2% max: -4.1% clap-rs-check avg: -2.9% min: -5.2% max: -1.1% hyper avg: -3.0% min: -5.1% max: -0.8% parser-check avg: -4.2% min: -4.9% max: -3.2% hyper-opt avg: -2.6% min: -4.9% max: -0.3% encoding-opt avg: -2.3% min: -4.6% max: -0.5% encoding avg: -2.5% min: -4.4% max: -0.6% issue-46449 avg: -2.3% min: -4.4% max: -1.8% issue-46449-opt avg: -1.7% min: -4.3% max: -0.9% clap-rs-opt avg: -1.6% min: -4.2% max: -0.2% serde-check avg: -1.4% min: -4.1% max: -0.2% clap-rs avg: -1.6% min: -3.9% max: -0.7% unify-linearly-check avg: -3.2% min: -3.7% max: -2.7% serde avg: -1.1% min: -3.5% max: -0.1% regression-31157-check avg: -2.6% min: -3.4% max: -1.6% helloworld-check avg: -2.5% min: -3.4% max: -0.6% serde-opt avg: -1.3% min: -3.3% max: -0.5% tokio-webpush-simple-check avg: -2.4% min: -3.2% max: -1.8% piston-image-check avg: -1.7% min: -3.2% max: -0.9% deeply-nested-opt avg: -1.5% min: -3.0% max: -0.6% deeply-nested-check avg: -1.9% min: -2.9% max: -0.4% deeply-nested avg: -1.9% min: -2.9% max: -1.2% syn-check avg: -1.8% min: -2.8% max: -0.6% coercions avg: -0.5% min: -2.8% max: 0.4% syn-opt avg: -0.9% min: -2.4% max: -0.1% syn avg: -1.1% min: -2.2% max: -0.3% parser-opt avg: -1.9% min: -2.1% max: -1.6% parser avg: -1.9% min: -2.1% max: -1.6% style-servo-check avg: -1.3% min: -2.0% max: -0.8% regression-31157-opt avg: -0.8% min: -2.0% max: 0.0% piston-image avg: -0.7% min: -1.8% max: -0.2% piston-image-opt avg: -0.6% min: -1.8% max: -0.0% regression-31157 avg: -1.0% min: -1.7% max: -0.3% html5ever-opt avg: -0.6% min: -1.5% max: -0.1% unify-linearly-opt avg: -1.3% min: -1.5% max: -1.1% unify-linearly avg: -1.3% min: -1.4% max: -1.2% tokio-webpush-simple-opt avg: -0.4% min: -1.2% max: -0.0% helloworld-opt avg: -1.0% min: -1.1% max: -0.6% helloworld avg: -1.0% min: -1.1% max: -0.7% inflate-opt avg: -0.3% min: -1.1% max: 0.1% html5ever-check avg: -0.6% min: -1.0% max: -0.3% inflate-check avg: -0.3% min: -1.0% max: -0.1% ``` </details>

bors · 2018-05-03T10:28:32Z

☀️ Test successful - status-appveyor, status-travis
Approved by: eddyb
Pushing 698b956 to master...

rust-highfive assigned jseyfried May 2, 2018

rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label May 2, 2018

kennytm reviewed May 2, 2018

View reviewed changes

kennytm added the T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. label May 2, 2018

kennytm added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels May 2, 2018

nnethercote force-pushed the escape_unicode branch from e8cf316 to bb18f41 Compare May 2, 2018 23:52

kennytm added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels May 2, 2018

kennytm reviewed May 2, 2018

View reviewed changes

nnethercote added 2 commits May 3, 2018 10:31

Remove parse::escape_default().

7a56360

str::escape_default() can be used instead.

nnethercote force-pushed the escape_unicode branch from bb18f41 to 7a56360 Compare May 3, 2018 01:12

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels May 3, 2018

bors merged commit 7a56360 into rust-lang:master May 3, 2018

nnethercote deleted the escape_unicode branch May 4, 2018 04:00

lowr mentioned this pull request Aug 5, 2022

fix: make concat! work with char rust-lang/rust-analyzer#12942

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use escape_default() for strings in LitKind::token(). #50391

Use escape_default() for strings in LitKind::token(). #50391

nnethercote commented May 2, 2018 •

edited by kennytm

Loading

kennytm May 2, 2018

nnethercote May 2, 2018

eddyb commented May 2, 2018

nnethercote commented May 2, 2018

kennytm May 2, 2018

nnethercote commented May 3, 2018

kennytm commented May 3, 2018

bors commented May 3, 2018

bors commented May 3, 2018

bors commented May 3, 2018

Use escape_default() for strings in LitKind::token(). #50391

Use escape_default() for strings in LitKind::token(). #50391

Conversation

nnethercote commented May 2, 2018 • edited by kennytm Loading

kennytm May 2, 2018

Choose a reason for hiding this comment

nnethercote May 2, 2018

Choose a reason for hiding this comment

eddyb commented May 2, 2018

nnethercote commented May 2, 2018

kennytm May 2, 2018

Choose a reason for hiding this comment

nnethercote commented May 3, 2018

kennytm commented May 3, 2018

bors commented May 3, 2018

bors commented May 3, 2018

bors commented May 3, 2018

nnethercote commented May 2, 2018 •

edited by kennytm

Loading