provide Olm as a natively linked module for riot-desktop

[ara4n]

No description provided.

[lampholder]

@ara4n what does this get us? I don't understand well enough to put any sort of priority on this :P

[ara4n]

it'd make crypto run ~50x faster on desktop, and make it more resistant to timing attacks. but given we need crypto to run at a usable speed on the web anyway the 50x speedup is probably not going to change much - it'd more be a case of using less cpu. it's prolly p3 e2e feature.

[richvdh]

I'd prioritise element-hq/element-web#2503 and matrix-org/olm#3 over this.

[uhoreg]

I've created a preliminary native module for olm, available at https://gitlab.com/uhoreg/olm-native. I haven't done much checking to make sure that it's correct, other than running the olm unit tests, but the results show a definite improvement in speed. On my computer:

transpiled JavaScript:

> jasmine-node test --verbose --junitreport --captureExceptions

têst1 -> { plaintext: 'têst1', message_index: 0 }
hot beverage: ☕ -> { plaintext: 'hot beverage: ☕', message_index: 1 }
☕ -> { plaintext: '☕', message_index: 2 }
têst1 -> têst1
hot beverage: ☕ -> hot beverage: ☕

megolm - 638 ms
    should encrypt and decrypt - 637 ms

olm - 1273 ms
    should encrypt and decrypt - 1273 ms

Finished in 1.916 seconds
2 tests, 11 assertions, 0 failures, 0 skipped

native library:

> jasmine-node test --verbose --junitreport --captureExceptions

têst1 -> { plaintext: 'têst1', message_index: 0 }
hot beverage: ☕ -> { plaintext: 'hot beverage: ☕', message_index: 1 }
☕ -> { plaintext: '☕', message_index: 2 }
têst1 -> têst1
hot beverage: ☕ -> hot beverage: ☕

megolm - 17 ms
    should encrypt and decrypt - 15 ms

olm - 5 ms
    should encrypt and decrypt - 4 ms

Finished in 0.027 seconds
2 tests, 11 assertions, 0 failures, 0 skipped

I ran the test a few times, and the numbers are fairly consistent. With the native version, megolm is about 37 times faster, and olm is about 250 times faster.

[MTRNord]

About packaging native stuff with electron: Check my spellcheck pr. I already added commands in there to recompile everything to glue it to the correct electron on packaging

[turt2live]

Pardon my thickness, but isn't this WASM?

[jryans]

I think this issue is suggesting we use Olm as a native binary library, such as *.dll / *.dylib / *.so and friends, with some JS bindings as an interface layer (instead of compiling the C code to WASM). That appears to be what https://gitlab.com/uhoreg/olm-native works towards.

[richvdh]

I don't think this is relevant to Element desktop any more.