Analyze JavaScript Bundle Size
"Which import added 300 kB" is answerable in two commands; guessing wrong wastes a day. Measure the parsed size, not the gzipped file listing.
Procedure
Record the baseline first:
du -sh dist/assets/*.js | sort -h | tail -20Emit a stats file from the bundler, then open the treemap:
npx webpack --profile --json > dist/stats.json npx webpack-bundle-analyzer dist/stats.jsonFor a source-accurate view, keep source maps in the build and map every chunk back to modules:
npx source-map-explorer 'dist/assets/*.js' --gzipSort by gzip size, then inspect the top three. A duplicate copy of a library (two React versions) shows as two adjacent nodes.
Check side-effect flags; a package without
"sideEffects":falsedefeats tree-shaking and drags whole barrels in.Replace a heavy leaf dependency only after measuring it: moment -> date-fns/dayjs, lodash ->
lodash-esnamed imports, axios ->fetch.Re-run step 3 and record the delta; a bundle action without a before/after number is not a result.
Pitfalls
- Reading the network panel's transfer size and calling it runtime cost; parse and compile time scale with the uncompressed size.
- Analyzing production builds with
mode: development, where nothing is minified and the numbers are 5x inflated. - Removing source maps to shrink the build, which blinds every future analysis.
- Chasing the largest chunk while a 40 kB synchronous module sits in the critical path and blocks first paint.
- Importing from a package root (
import {x} from 'lib') when only one function is needed, pulling the whole barrel.
Verification
npx source-map-explorer 'dist/assets/*.js' --gzip --json | jq '.results[0].totalBytes'
The total gzipped bytes must be at or below the recorded baseline minus the intended reduction. Report the metric name, the before/after bytes, and the module that moved.