Findings on non-global require() in CLI commands

[ryanblock]

Begin is (currently) a Node.js app compiled with pkg, and I wanted to ensure it felt as fast as any C/C++/Rust/etc. CLI tool.

In the course of building the CLI I paid careful attention to its performance characteristics, and observed that the idiomatic usage of global require() carried a fairly massive penalty as larger dependencies (such as @architect/sandbox) were included. While @architect/sandbox isn't itself all that heavy, it does require aws-sdk, which is >50MB.

The basic principle is: less code loaded means faster runs (in this case by 3-10x). That's it, no magic.

Here are a few anecdotal examples, which I encourage you to verify for yourself! These are run on a 2019 2.4GHz Core i9 MacBook Pro:

begin version (not compiled), Sandbox not required globally

✓ [12:23:54] (╯ಠ益ಠ)╯︵~/git/begin/cli [main]
⮑  time node . version
Begin 0.0.0

real    0m0.066s
user    0m0.053s
sys     0m0.014s

begin version (not compiled), global require of Sandbox

✓ [12:24:01] (╯ಠ益ಠ)╯︵~/git/begin/cli [main]
⮑  time node . version
Begin 0.0.0

real    0m0.652s
user    0m0.394s
sys     0m0.125s

begin version (compiled), Sandbox not required globally

✓ [12:28:22] (╯ಠ益ಠ)╯︵~/git/begin/cli [main]
⮑  time bin/begin version
Begin 0.0.0

real    0m0.081s
user    0m0.069s
sys     0m0.014s

begin version (compiled), global require of Sandbox

✓ [12:28:55] (╯ಠ益ಠ)╯︵~/git/begin/cli [main]
⮑  time bin/begin version
Begin 0.0.0

real    0m0.340s
user    0m0.379s
sys     0m0.045s