[wasm] Turn Wasm exception handling on by default and use WasmEnableExceptio…

[lewing]

…nHandling as the default value for WasmEnableSIMD

[radical]

WBT/node tests failing with:

Application arguments: --run async_main_with_args_Release_False.dll abc foobar
MONO_WASM: instantiate_wasm_module() failed: {}
MONO_WASM: Stacktrace:

CompileError: WebAssembly.instantiate(): unexpected section <Tag> (enable with --experimental-wasm-eh) @+15097
CompileError: WebAssembly.instantiate(): unexpected section <Tag> (enable with --experimental-wasm-eh) @+15097
exiting due to exception: CompileError: WebAssembly.instantiate(): unexpected section <Tag> (enable with --experimental-wasm-eh) @+15097

[lewing]

WBT/node tests failing with:

Application arguments: --run async_main_with_args_Release_False.dll abc foobar
MONO_WASM: instantiate_wasm_module() failed: {}
MONO_WASM: Stacktrace:

CompileError: WebAssembly.instantiate(): unexpected section <Tag> (enable with --experimental-wasm-eh) @+15097
CompileError: WebAssembly.instantiate(): unexpected section <Tag> (enable with --experimental-wasm-eh) @+15097
exiting due to exception: CompileError: WebAssembly.instantiate(): unexpected section <Tag> (enable with --experimental-wasm-eh) @+15097

yeah we need to enable the feature for all the host runtimes

[radekdoulik]

curious: why to base the default SIMD value on EH value?

[lewing]

because it is much less of an improvement without EH and EH is more broadly compatible/deployed than SIMD so if you are explicitly disabling EH you probably want to disable SIMD too, whereas if you were disable SIMD for compat reasons you may not need to disable EH. That was my thought at least

[radekdoulik]

EH is more broadly compatible/deployed than SIMD

only in Safari. Chrome, Firefox and nodejs have SIMD support longer than EH. (https://webassembly.org/roadmap/)

I think the new behavior is a bit confusing. Maybe we could introduce a new property, like WasmEnableAdvancedFeatures or WasmEnableNextGen to switch the defaults of SIMD, EH and other non 1.1 features. What do you think?

[kg]

Looks fine other than radek's comment

[lewing]

@radekdoulik should we change the benchmarks to explicitly set these settings based on their name/intension?