Reduce WASM build size

[Thoralf-M]

Description of change

With these changes the wasm file size should be reduced from 4,26 MB to 2,59 MB

Type of change

• Enhancement (a non-breaking change which adds functionality)

How the change has been tested

Executed an example

[cycraig]

Thanks for the PR!

Short version

• we lose debug symbols in stack traces but the reduction in size is worth it,
• there is a slight reduction in runtime performance but it's still more than "fast enough" in my opinion,
• enabling full link-time-optimisation would help slightly,
• there may be issues with wasm-opt binaries not being pre-built for M1 Mac,

Overall looks good to me! Would merge as-is.

Long version

I wonder if there was any reason for wasm-opt = false in the first place? It's been set from the inception of the wasm bindings:

499db7c#diff-c3457202cbb77c82da5b8f2b6dcdc50b85bd7e343832927909ff1f0049b4f1bdR15

# TODO: Discover this:
[package.metadata.wasm-pack.profile.release]
wasm-opt = false

Perhaps there was/is an issue with wasm-opt lacking binaries for certain targets/architectures? Most likely Macintosh (M1 chip or not), similar to wasm-pack in #415. The npm package is built and published using Ubuntu in GitHub Actions, so releases are not a problem, but for local development we could introduce build-dev/run-dev steps to avoid any issues with wasm-opt and debug symbols.

Size Comparison

Out of interest, I tested the effects of different compilation options on both size and runtime performance. I included different combinations of lto options since it was recommended at https://rustwasm.github.io/book/reference/code-size.html

The options tested are:

• wasm-opt ∈ [false, true],
• opt-level ∈ [3 (default), 's', 'z'],
• lto ∈ [false, "thin", true].

The compressed and unpacked sizes are gathered from npm pack. Build time is measured from a clean wasm target directory with cargo clean && rm -r node && rm -r web && (Measure-Command { npm run build | Out-Default }).ToString().

	wasm-opt	opt-level	lto	wasm size (MB)	compressed size (MB)	unpacked size (MB)	build time (s)
1 (current)	false	3	false	4.46	2.7	9.3	56
2	true	3	false	3.57	2.3	7.5	93
3	true	3	"thin"	3.57	2.3	7.5	90
4	true	3	true	3.57	2.3	7.5	91
5 (this PR)	true	s	false	2.71	1.9	5.8	62
6	true	s	"thin"	2.71	1.9	5.8	65
7	true	s	true	2.72	1.9	5.8	64
8	true	z	false	2.61	1.8	5.6	61
9	true	z	"thin"	2.61	1.8	5.6	62
10	true	z	true	*2.60	1.8	5.6	67

Highlights:

• 'z' reduces the size by another ~100KB compared to 's'
• lto has a negligible effect on wasm size, ~10KB at most
• wasm-opt adds ~35s to the build time, but changing the opt-level offsets this (compilation time was measured in case we have any significant increase in CI time, it doesn't really matter how long release builds take compared to package size and speed)

Performance Comparison

Unfortunately we don't have proper benchmarks for the wasm bindings: there are issues with running criterion.rs for wasm and the runtime of examples varies wildly due to network calls and remote proof-of-work. So I settled on the following micro-benchmark to avoid wasting too much time:

benchmark.js:

const LIMIT = 10000;

// Time generating a new Ed25519 keypair and document, signing and verifying it
console.time('generate, sign, verify document');
for (let i = 0; i < LIMIT; i++) {
    let key = new KeyPair(KeyType.Ed25519);
    let doc = new Document(key, "main");
    doc.sign(key);
    console.assert(doc.verify())
}
console.timeEnd('generate, sign, verify document');

npm run build && node benchmark/benchmark.js:

	wasm-opt	opt-level	lto	mean time (ms)	total time (s)
1 (current)	false	3	false	0.862	8.631
2	true	3	false	0.841	8.420
3	true	3	"thin"	0.841	8.417
4	true	3	true	0.857	8.580
5 (this PR)	true	s	false	1.041	10.415
6	true	s	"thin"	0.989	9.898
7	true	s	true	0.988	9.882
8	true	z	false	1.077	10.770
9	true	z	"thin"	1.087	10.872
10	true	z	true	1.065	10.658

Highlights:

• 's' increases runtime by ~20% but overall still ~1ms so it's fine,
• lto = true very slightly reduces runtime for 's' and 'z'

Debug Symbols

Lastly, it's important to note that the size reduction from wasm-opt is in large part due to stripping out the debug info symbols. This removes Rust function names from error stack traces, but the Javascript wrapper entry-point is still output so it's probably good enough. The "-g" flag can be passed to wasm-opt to preserve debug symbols but then there is almost no reduction in package size.

Example Stack Trace

Generate an error by running

let key = new KeyPair(KeyType.Ed25519);
let doc = new Document(key, "Invalid;Network{}Name!");

1. Current wasm-opt = false:

InvalidNetworkName: Invalid Network Name
    at module.exports.__wbg_new_342a24ca698edd87 (C:\Work\iota\tmp\identity.rs\bindings\wasm\node\identity_wasm.js:3981:15)
    at js_sys::Error::new::ha2f916f5a59a175d (<anonymous>:wasm-function[5350]:0x31b609)
    at identity_wasm::error::wasm_error::h62760bc7e7cb9ef0 (<anonymous>:wasm-function[1637]:0x27a46c)
    at identity_wasm::did::wasm_document::WasmDocument::new::h513b33a29593d84d (<anonymous>:wasm-function[677]:0x1d5461)
    at document_new (<anonymous>:wasm-function[2851]:0x2e4107)
    at new Document (C:\Work\iota\tmp\identity.rs\bindings\wasm\node\identity_wasm.js:1373:24)
    at Object.<anonymous> (C:\Work\iota\tmp\identity.rs\bindings\wasm\benchmark\benchmark.js:53:15)

2. This PR wasm-opt = true, function names are stripped:

InvalidNetworkName: Invalid Network Name
    at module.exports.__wbg_new_342a24ca698edd87 (C:\Work\iota\tmp\identity.rs\bindings\wasm\node\identity_wasm.js:3981:15)
    at <anonymous>:wasm-function[4319]:0x2b617b
    at <anonymous>:wasm-function[1566]:0x237218
    at <anonymous>:wasm-function[665]:0x1a79c3
    at document_new (<anonymous>:wasm-function[2615]:0x28d9a0)
    at new Document (C:\Work\iota\tmp\identity.rs\bindings\wasm\node\identity_wasm.js:1373:24)
    at Object.<anonymous> (C:\Work\iota\tmp\identity.rs\bindings\wasm\benchmark\benchmark.js:53:15)

3. Preserve debug symbols with wasm-opt = ["-g"]:

InvalidNetworkName: Invalid Network Name
    at module.exports.__wbg_new_342a24ca698edd87 (C:\Work\iota\tmp\identity.rs\bindings\wasm\node\identity_wasm.js:3966:15)
    at js_sys::Error::new::h105995e043d6d4ab (<anonymous>:wasm-function[8784]:0x23648e)
    at identity_wasm::error::<impl core::convert::From<identity_wasm::error::WasmError> for js_sys::Error>::from::h0ec0dcb249a1399f (<anonymous>:wasm-function[4613]:0x2052e0)
    at identity_wasm::error::wasm_error::h679364300396755a (<anonymous>:wasm-function[3832]:0x1ee266)
    at identity_wasm::did::wasm_document::WasmDocument::new::ha412fd768f0c1591 (<anonymous>:wasm-function[994]:0x13497f)
    at document_new (<anonymous>:wasm-function[1504]:0x16a021)
    at new Document (C:\Work\iota\tmp\identity.rs\bindings\wasm\node\identity_wasm.js:1373:24)
    at Object.<anonymous> (C:\Work\iota\tmp\identity.rs\bindings\wasm\benchmark\benchmark.js:53:15)

Unfortunately using "-g" to retain/add debug symbols barely reduces the size at all:

	wasm-opt	opt-level	wasm size (MB)	package size (MB)	unpacked size (MB)
1 (current)	false	3	4.46	2.7	9.3
5 (this PR)	true	s	2.71	1.9	5.8
8	true	z	2.61	1.8	5.6
11	["-g"]	s	3.93	2.3	8.6
12	["-g"]	z	4.21	2.4	8.8

[Thoralf-M]

Thanks for this extensive overview of the different optimization levels!

[cycraig]

Enabled lto and added a build-dev task to bypass wasm-opt and retain symbols for debugging locally following internal discussions.

@eike-hass Please sanity-check the build-dev task.

[eike-hass]

Enabled lto and added a build-dev task to bypass wasm-opt and retain symbols for debugging locally following internal discussions.

@eike-hass Please sanity-check the build-dev task.

Tested on WSL 2 / Ubuntu 20, LGTM