Do not regenerate api object that does not change

[paololim]

The Dynamic team is using siwe-parser in our codebase in production, and noticed that ParsedMessage(text) was causing unnecessary latency in our application. When we took a closer look its implementation, we noticed that the bulk of the latency was from creating a new apgApi:

const api = new apgApi(GRAMMAR);

Doing this in the constructor of ParsedMessage is unnecessary because new apgApi(GRAMMAR) would never actually change and that GRAMMAR is a static const declared earlier in the file. This means every time we instantiate the ParsedMessage class, we generate a new and unnecessary apgApi.

This PR removes the expensive API generation from ParsedMessage's constructor and make sure we only generate the API once.

siwe-parser's tests have a bunch of test cases they evaluate against to check that the parser works. Locally, I added basic time duration that looks like:

const start = new Date().getTime();
const parsedMessage = new ParsedMessage(test.message); // this line was already in parser.test.ts
const duration = new Date().getTime() - start;
console.log(`test duration[${test_name}]: ${new Date().getTime() - start}`)

Here are the results collected from running the test BEFORE I made the change (this was running on a M1 Max):

test duration[couple of optional fields]: 53
test duration[no optional field]: 35
test duration[timestamp without microseconds]: 29
test duration[domain is RFC 3986 authority with IP]: 26
test duration[domain is RFC 3986 authority with userinfo]: 25
test duration[domain is RFC 3986 authority with port]: 21
test duration[domain is localhost authority with port]: 21
test duration[domain is RFC 3986 authority with userinfo and port]: 22
test duration[no statement]: 21
test duration[domain ipv6]: 22
test duration[uri ipv6]: 22
test duration[uri ipv4]: 21
test duration[uri with port]: 21
test duration[uri ipv4 query params and fragment]: 21
test duration[chainId not 1]: 21
test duration[recovery byte starting at 0]: 21

And here are the results collected after the change was made:

test duration[couple of optional fields]: 11
test duration[no optional field]: 2
test duration[timestamp without microseconds]: 2
test duration[domain is RFC 3986 authority with IP]: 1
test duration[domain is RFC 3986 authority with userinfo]: 1
test duration[domain is RFC 3986 authority with port]: 1
test duration[domain is localhost authority with port]: 2
test duration[domain is RFC 3986 authority with userinfo and port]: 2
test duration[no statement]: 0
test duration[domain ipv6]: 1
test duration[uri ipv6]: 2
test duration[uri ipv4]: 0
test duration[uri with port]: 0
test duration[uri ipv4 query params and fragment]: 0
test duration[chainId not 1]: 0
test duration[recovery byte starting at 0]: 0

As you can see, the duration for instantiating the ParsedMessage class goes down significantly after the change - and this would translate to much lower latency in our application and the endpoint that uses this.

[paololim]

Hi there! I just wanted to follow-up on this PR. Anything we can do to help with the review? Thank you!

[w4ll3]

Thanks for you PR, sorry for the late reply as well. We'll be looking into it.

[sbihel]

Could you just rebase your PR before I merge please?

[paololim]

hi @sbihel @w4ll3 - thanks for taking a look! i have rebased the commit off of upstream/main, and should be correct now. let me know if you need anything else.

And also made sure tests pass locally:

[/Users/paolo/Dyn/siwe:faster_parser]☕ npm run test
> siwe@2.0.5 test
> jest
 PASS  packages/siwe-parser/lib/parsers.test.ts
 PASS  packages/siwe/lib/client.test.ts (14.525 s)
....
....
....
Test Suites: 2 passed, 2 total
Tests:       209 passed, 209 total
Snapshots:   0 total
Time:        14.825 s
Ran all test suites in 2 projects.

[paololim]

@skgbafa @sbihel Thanks for reviewing and merging! When can we expect the new version to be published to NPM?

[w4ll3]

I can't give you an exact date right now, but we have a new release planned already and this changes will be included.