Local/Remote Execution environment

[partouf]

First working version

This is part 1 of many for #6379

Changed in this PR:

• Move User Executable Execution into it's own independent thing
• Stored some extra things in buildResult so execution can work without knowledge of the compiler
• Be able to use a Hash and some JSON to execute

This so that later:

• We can add and use a Remote Execution Environment that pushes to a remote machine instead of directly executing (queues etc, see [REQUEST]: Remote execution on different architectures #6379)
• The remote machine and architecture can run the executable and return the results

Will comment out the API Endpoint before merging of course

[partouf]

Took more refactoring then I had hoped, but I'm happy with the end-result

[partouf]

How I tested this:

1. Build executable with regular CE with debug on
2. Copy paste the hash for the cached executable (123abc_exec hash)
3. Use the hash on script below:

#!/bin/sh

get_post_json() {
	cat <<'EOF'
{
	"ExecutionParams": {
		"args": ["Hello", "args!"],
		"stdin": "Hello stdin!",
		"runtimeTools": [
			{
				"name": "env",
				"options": [
					{
					"name": "PETE",
					"value": "Hello123"
					}
				]
			},
			{
				"name": "heaptrack",
				"options": [
					{
						"name": "details",
						"value": "stderr"
					},
					{
						"name": "graph",
						"value": "yes"
					}
				]
			}
		]
	}
}
EOF
}

POSTJSON=$(get_post_json)

echo $POSTJSON | jq

curl -s "http://127.0.0.1:10240/api/localexecution/$1" -H "Accept: application/json" -H "Content-type: application/json" -d "$POSTJSON" | jq

[mattgodbolt]

...will comment out the API endpoint

Maybe make it a configuration option (default off, naturally).

[mattgodbolt]

Remarkably little change to get this I see (mostly code motion), nice work. Discussing more in person.

[mattgodbolt]

Just to understand "local execution" is what in this instance? Is this local as in "on the local server itself" - the thing you said you'd comment out? I can see if being useful for testing, so another +1 on keep-it-behind-an-option.

[partouf]

Execute on this instance yes, definitely just for testing. And I can see an option being useful, but I don't want users to accidentally turn it on, as nothing in this function is sanitized...

[mattgodbolt]

Understood, but we have plenty of options that are "buyer beware" for debug etc. I think having it in an option makes sense, else how might we test it?

[mattgodbolt]

Nice

[partouf]

Breaks Heaptrack Graph in Cmake + Execution https://godbolt.org/staging/z/zz6M7TWqx
(works on prod)

[partouf]

F# execution broken https://godbolt.org/staging/z/W3nazK8oc