fix(gatsby): don't force leading slash for external paths in routes manifest

[jenae-janzen]

Fixes: https://linear.app/netlify/issue/FRA-20/external-redirects-have-a-forced-slash-added

If route.path begins with http(s)://, we shouldn't add a leading slash to it.

Description

Documentation

Tests

Related Issues

Fixes FRA-20

[jenae-janzen]

🤔 Not sure why the tests are failing, I don't see that they're related to this change:

[MarcL]

Awesome! Can we add a unit test for this code @jenae-janzen?

The other unit tests are here:
https://github.com/gatsbyjs/gatsby/blob/master/packages/gatsby/src/utils/adapter/__tests__/manager.ts#L47

cc @pieh for any guidance if needed.

[pieh]

🤔 Not sure why the tests are failing, I don't see that they're related to this change:

Some tests are quite flaky :( - the failing test suites are not even using/testing any of code changes (I did just restart them)

cc @pieh for any guidance if needed.

gatsby/packages/gatsby/src/utils/adapter/__tests__/fixtures/state.ts

Lines 67 to 73 in 3af35ae

	const redirects: IGatsbyState["redirects"] = [{
	fromPath: '/old-url',
	isPermanent: true,
	ignoreCase: true,
	redirectInBrowser: false,
	toPath: '/new-url'
	}]

this would need some redirects added that would have external fromPath and then assertions/ tests done in https://github.com/gatsbyjs/gatsby/blob/master/packages/gatsby/src/utils/adapter/__tests__/manager.ts#L47

[jenae-janzen]

Any performance concerns using a regex here, i.e. if a site has a lot of redirects?

[pieh]

doing route.path.startsWith('http://') || route.path.startsWith('https://') is faster than using that exact regex (tho doesn't cover case insensitivity) - so I'll adjust to use this.

"RegExp x 1,614,517 ops/sec ±0.62% (96 runs sampled)"
"startWith x 5,467,939 ops/sec ±0.17% (96 runs sampled)"
"Fastest is startWith"

Note here that it seems like initiating new regexp on each addRoute call is siginificant portion of overhead it seems - after moving regexp to be inited once earlier we get

"RegExp x 2,410,360 ops/sec ±0.99% (95 runs sampled)"
"startWith x 5,037,209 ops/sec ±3.31% (93 runs sampled)"
"Fastest is startWith"

so 2 startWith calls is still faster then regexp test (the startWith benchmark case was using 2 calls, not 1, so this will be ~twice as fast)

I'll adjust to use startWith and get PR merged

[pieh]

Released in gatsby@5.12.8