Homebrew Bottle Size Analysis

Analyzing the size of Homebrew formulae bottles

Intent

The Homebrew formula JSON API does not provide package size information for bottles¹. I aim to retrieve package sizes regularly in order to build a database of (package, version, bottle_arch) -> size pairs for future analysis. This analysis could capture:

Package growth over time
Total estimated archive size
Spikes in package size indicating significant changes warranting further inspection
Expired or broken package URL for rarely-updated formulae with rarely-downloaded bottles
Packages to target for size optimization, from individual relief to humanity-scale savings.

This is currently mostly an experiment in using simple CLI tools like Make and curl to do some data engineering and science that has the above useful implications.

Current principles

KISS, to the level of probably dumb.
Use CLI tools as much as possible; keep code to a minimum.
Anything that can be installed via Homebrew is fair to use.
Prioritize concurrency using simple tools such as Make -j, xargs, parallel, fd, ripgrep, etc.
Rebuilding the database from scratch means losing data, so avoid that.

Usage

make formula.json  # get the data file
make urls          # split it out
make sizes         # get the sizes

Architecture

stateDiagram-v2
   Formulajson : Homebrew API \n formula.json
   Urls : One URL file per URL
   Database : Database (Unspecified Format)

    [*] --> Formulajson : retrieve latest database
    Formulajson --> Urls : extract triplets, write URL files
    state fork_state <<fork>>

    state join_state <<join>>
    Urls --> fork_state : list URL files
    fork_state --> HTTPRequest1 : retrieve package size
    fork_state --> HTTPRequest2 : through HEAD requests
    fork_state --> HTTPRequestN : to all URLs in files
    HTTPRequest1 --> Sizes1 : write size file
    HTTPRequest2 --> Sizes2 : write size file
    HTTPRequestN --> SizesN : write size file
    Sizes1 --> join_state
    Sizes2 --> join_state
    SizesN --> join_state
    join_state --> Database
    Database --> [*]

    note left of fork_state
      One size file per retrieved URL
    end note

Performance notes

It takes around 80 minutes to run for me two requests at a time in order not to trigger some kind of speed limit at my ISP level ².

You can check the counts of urls and size files by running something like this:

fd .url data | wc -l
fd .size data | wc -l

If the numbers are the same, you've got the data for the current formula.json.

A bottle is a pre-packaged archive of a formula available in Homebrew. See https://docs.brew.sh/Bottles for more information. ↩
It's not ghcr.io rate-limiting me. My gateway is working fine but my ISP drops the upstream connection. It's probably some kind of DDOS protection at the DNS level. See notes.txt for ways I might get around this since curl does a DNS lookup every time it launches. ↩