perf(autolinking)!: prioritise files with *Package.java|*Package.kt in getPackageClassName
#2384
Conversation
|
|
||
| function getClassNameMatches(files: string[], folder: string) { | ||
| return files | ||
| .map((filePath) => fs.readFileSync(path.join(folder, filePath), 'utf8')) |
There was a problem hiding this comment.
I wonder if streaming is faster than reading the whole file every time: https://tangledeveloper.medium.com/how-to-read-file-line-by-line-efficiently-in-node-js-8125fe5bd185
There was a problem hiding this comment.
worth to measure definitely, these declarations are usually somewhere at the top of the file so it hints some potential
There was a problem hiding this comment.
Replacing glob.sync above with glob.stream might also be worth looking into, but making everything async might turn this into a very big PR 😛
| .map(matchClassName) | ||
| .filter((match) => match); | ||
| if (!packages.length) { | ||
| files = getMainActivityFiles(folder, false); |
There was a problem hiding this comment.
maybe there's scope to exclude previously matched files from this list not to read same files twice
There was a problem hiding this comment.
nice! good catch
cortinico
left a comment
cortinico
left a comment
There was a problem hiding this comment.
Great improvement 👍 All good on my end but wait for another CLI reviews before merging this (also because I believe this is not E2E tested)
szymonrybczak
force-pushed
the
perf/autolinking/smaller-glob-schema
branch
from
e93df86 to
6806649
Compare
| let files = getMainActivityFiles(folder); | ||
| let packages = getClassNameMatches(files, folder); | ||
|
|
||
| const packages = files | ||
| .map((filePath) => fs.readFileSync(path.join(folder, filePath), 'utf8')) | ||
| .map(matchClassName) | ||
| .filter((match) => match); | ||
| if (!packages.length) { | ||
| files = getMainActivityFiles(folder, false); |
There was a problem hiding this comment.
A bit late to the game but I wonder if globbing for **/*.{java,kt} once and filtering in-memory instead of hitting the disk twice is faster for bigger projects.
There was a problem hiding this comment.
@szymonrybczak keen to measure it and sending a followup if that helps with perf? :)
There was a problem hiding this comment.
Hm, I gave it a shot but the improvement won't be big because after my checks the amount of libraries that needs to fallback are very small.
Now:
❯ hyperfine "npx react-native config --platform android"
Benchmark 1: npx react-native config --platform android
Time (mean ± σ): 2.070 s ± 0.264 s [User: 1.049 s, System: 0.607 s]
Range (min … max): 1.916 s … 2.814 s 10 runs
With executing glob once and filtering in-memory:
❯ hyperfine "npx react-native config --platform android"
Benchmark 1: npx react-native config --platform android
Time (mean ± σ): 2.096 s ± 0.190 s [User: 1.078 s, System: 0.620 s]
Range (min … max): 1.931 s … 2.572 s 10 runs
Summary:
This Pull Request changes schema that looks for files to check to find class name, after my tests in various application the match results were included almost in every library in files with
Packageat the end. But our previous logic was going through every file in library with.ktand.javaextension until it found file withimplements ReactPackage, some libraries can have 100+ files and we were reading them until we matched relevant Regex schema 😅I also added a fallback which checks files with
*.ktand*.javaextensions if it couldn't findimplements ReactPackagein file with*Packageat the end, which ensures us that we don't introduce any breaking change. The output ofconfigcommand stays the same.Performance improvement
I did a test in Expensify app, which is an open-source app with 50 libraries containing native code. Before my changes, we were reading 490 files to find the class name. Right now, we only check 79 (still there's a room for improvements).
Test Plan:
The
dependenciesfield should be the same as before.Checklist