meta: remove let from for loops

[MylesBorins]

Checklist

• make -j8 test (UNIX), or vcbuild test nosign (Windows) passes
• commit message follows commit guidelines

Affected core subsystem(s)

lib

Description of change

This is a known de-opt. It may not be 100% necessary in all cases but it
seems like a decent enough idea to avoid it.

If individuals are interested we can create a linting rule for this

[mscdex]

LGTM if CI is ok with it: https://ci.nodejs.org/job/node-test-pull-request/4343/

[mscdex]

Looks like there's a variable name conflict between a var and a const.

[MylesBorins]

fixed

new ci: https://ci.nodejs.org/job/node-test-pull-request/4344/

[ChALkeR]

Does this have any effect on the benchmarks in master?

[MylesBorins]

windows failure was a timeout
https://ci.nodejs.org/job/node-test-binary-windows/4129/RUN_SUBSET=0,VS_VERSION=vs2015,label=win2012r2/tapTestReport/test.tap-238/

likely infra related

[mscdex]

Still LGTM.

[lpinca]

LGTM

[jbergstroem]

LGTM

[MylesBorins]

would someone from @nodejs/benchmarking help me run some tests to see if this actually makes any difference?

[imyller]

LGTM

[AndreasMadsen]

would someone from @nodejs/benchmarking help me run some tests to see if this actually makes any difference?

I doubt it makes a practical difference for some of these changes. For example calling stream.unpipe() is quite rare. So I just benchmarked the punycode changes:

# compile master 
git checkout master
./configure
make -j8
cp ./node ./node-master

# compile pr 8873
apply-pr 8873
make -j8
cp ./node ./node-pr8873

# benchmark
node benchmark/compare.js
  --old ./node-master --new ./node-pr8873
  --filter punycode -- net | tee punycode-benchmark.csv
cat punycode-benchmark.csv | Rscript ./benchmark/compare.R

                                                          improvement significant      p.value
val="belgië.icom.museum" n=1024 method="icu"                  -0.43 %             2.903102e-01
val="belgië.icom.museum" n=1024 method="punycode"              5.02 %         *** 5.920579e-12
val="českárepublika.icom.museum" n=1024 method="icu"          -0.08 %             8.350579e-01
val="českárepublika.icom.museum" n=1024 method="punycode"      7.01 %         *** 1.656402e-05
val="éire.icom.museum" n=1024 method="icu"                    -0.30 %             6.925716e-01
val="éire.icom.museum" n=1024 method="punycode"                5.72 %         *** 1.383591e-06
val="ísland.icom.museum" n=1024 method="icu"                   1.48 %             1.855450e-01
val="ísland.icom.museum" n=1024 method="punycode"              6.12 %         *** 3.057111e-07
val="magyarország.icom.museum" n=1024 method="icu"            -1.11 %             2.709653e-01
val="magyarország.icom.museum" n=1024 method="punycode"        6.48 %         *** 2.493271e-05
val="österreich.icom.museum" n=1024 method="icu"              -1.36 %             3.925142e-01
val="österreich.icom.museum" n=1024 method="punycode"          6.79 %         *** 1.343576e-05
val="ελλάδα.icom.museum" n=1024 method="icu"                   0.48 %             7.566853e-01
val="ελλάδα.icom.museum" n=1024 method="punycode"              5.80 %         *** 5.955543e-04
val="κυπρος.icom.museum" n=1024 method="icu"                  -1.55 %             3.175751e-01
val="κυπρος.icom.museum" n=1024 method="punycode"              4.83 %          ** 3.246308e-03
val="беларусь.icom.museum" n=1024 method="icu"                -0.23 %             7.148822e-01
val="беларусь.icom.museum" n=1024 method="punycode"            8.42 %         *** 6.554571e-04
val="българия.icom.museum" n=1024 method="icu"                 3.27 %             1.822171e-01
val="българия.icom.museum" n=1024 method="punycode"           10.95 %         *** 3.395971e-06
val="איקו״ם.ישראל.museum" n=1024 method="icu"                 -2.01 %             3.947640e-01
val="איקו״ם.ישראל.museum" n=1024 method="punycode"             5.00 %          ** 1.350123e-03
val="افغانستا.icom.museum" n=1024 method="icu"                -0.13 %             7.612569e-01
val="افغانستا.icom.museum" n=1024 method="punycode"            3.29 %           * 3.289352e-02
val="الأردن.icom.museum" n=1024 method="icu"                  -2.24 %             2.582112e-01
val="الأردن.icom.museum" n=1024 method="punycode"              4.67 %           * 1.154801e-02
val="الجزائر.icom.museum" n=1024 method="icu"                 -0.02 %             9.596995e-01
val="الجزائر.icom.museum" n=1024 method="punycode"             7.99 %         *** 2.736761e-04
val="القمر.icom.museum" n=1024 method="icu"                    0.47 %             2.381307e-01
val="القمر.icom.museum" n=1024 method="punycode"               5.69 %         *** 2.288273e-06
val="ايران.icom.museum" n=1024 method="icu"                   -0.33 %             8.802176e-01
val="ايران.icom.museum" n=1024 method="punycode"               2.77 %           * 1.163676e-02
val="تشادر.icom.museum" n=1024 method="icu"                   -0.11 %             7.246581e-01
val="تشادر.icom.museum" n=1024 method="punycode"               8.00 %         *** 3.214189e-05
val="مصر.icom.museum" n=1024 method="icu"                     -1.09 %             2.214773e-01
val="مصر.icom.museum" n=1024 method="punycode"                 6.64 %         *** 1.282843e-04
val="भारत.icom.museum" n=1024 method="icu"                    -0.46 %             3.967156e-01
val="भारत.icom.museum" n=1024 method="punycode"                5.03 %          ** 1.413984e-03
val="বাংলাদেশ.icom.museum" n=1024 method="icu"                -1.03 %             4.592365e-01
val="বাংলাদেশ.icom.museum" n=1024 method="punycode"            5.40 %           * 1.164211e-02
val="中国.icom.museum" n=1024 method="icu"                     3.20 %             5.088870e-02
val="中国.icom.museum" n=1024 method="punycode"                7.80 %         *** 2.616430e-06
val="日本.icom.museum" n=1024 method="icu"                     1.02 %             1.124426e-01
val="日本.icom.museum" n=1024 method="punycode"                4.01 %         *** 6.611998e-06

As you can see, the changes makes about a 5% statistically significant improvement in all punycode cases. If I didn't want the icu results I could have added --set method=punycode to compare.js.

PS: I'm not a @nodejs/benchmarking member, they may have a more historic (not just master) perspective.

[gibfahn]

@thealphanerd Looks like this has caused a couple of linter failures:

$ make lint              
./node tools/eslint/bin/eslint.js --cache --rulesdir=tools/eslint-rules \
      benchmark lib test tools

/Users/gib/dev/node/lib/_stream_readable.js
  664:30  error  'i' is constant         no-const-assign
  670:9   error  'i' is already defined  no-redeclare

✖ 2 problems (2 errors, 0 warnings)

make: *** [jslint] Error 1

[MylesBorins]

@gibfahn ah noodles. PR incoming

[mscdex]

Also for future reference (I missed this), lib: would have been a better subsystem name instead of meta:.

[MylesBorins]

I'm planning to force push to fix the problem. I can change the label when doing so

[MylesBorins]

force pushed with 2e568d9 to fix linting problem. Sorry about that, not sure how it regressed

[addaleax]

Just leaving the note here that if this lands in v4, it needs to do so together with #9171.

[ofrobots]

This just landed upstream in V8 5.7: https://bugs.chromium.org/p/v8/issues/detail?id=5666#c10. Starting in Node.js v8.x, we will no longer need to avoid let/const. \o/

[MylesBorins]

@Trott we should remove the lint rule that we have in V8... what is the best way to stage something like this until then

[Trott]

@thealphanerd Not sure. Maybe open a PR and label it with all the appropriate dont-land-on labels?

[Trott]

The above suggestion runs the risk of PRs landing on master that won't lint in LTS, but I don't think that will be too much of a problem in practice. We don't get a lot of let-in-loops PRs for the lib dir and when we do, reviewers usually call them out.

[mscdex]

Even though const/let will soon be handled by TurboFan, how does that currently stack up against just using var (which I assume still "goes through" Crankshaft)?

@ofrobots ?

[ofrobots]

Unless there is another reason to go through TF (e.g. try-catch, other ES6 features), the default would still be Crankshaft.

However, by the time this ships with Node.js 8.x, the default pipeline used by V8 would be quite different with Ignition and TurboFan being significantly more prominent. Exact details aren't locked down just yet.

I personally don't expect var, let and const to perform any different – but these don't exist in a vacuum. These are used in the context of other code and if the optimizer used to compile this surrounding code is changed, then you're comparing the capabilities of TF and CS on the surrounding code, not just var and let.

Try it out and report back if see unexpected performance results.