Install swift-help for integration tests. Suppress .noDriver options from being emitted by swift-help.

[cltnschlosser]

Fixes: Driver/help.swift

Also replaces deprecated createSymlink calls with localFileSystem.createSymbolicLink

[owenv]

Thanks!

[DougGregor]

@swift-ci test

[owenv]

LGTM!

[cltnschlosser]

@swift-ci test

[owenv]

@swift-ci test

[cltnschlosser]

@owenv -emit-module-doc-path, -index-ignore-stdlib, -emit-module-source-info are (seemingly incorrectly) marked .noDriver. Those are just the ones caught by the unit tests. I'll see if I can do a scan for any other .noDriver options that are used in the driver. Best solution is to update the options in c++ driver?

[cltnschlosser]

Okay, found 3 categories of .noDriver options that are used.

Definitely checked for and used:

.emitDependenciesPath
.emitModuleDocPath
.emitModuleDoc
.emitModuleSourceInfo
.enableCxxInterop

Technically supported, sets the compiler output type to pch, but I don't know if it actually works:

.emitPch

Erased when parsedOptions.hasArgument(.indexFile) && parsedOptions.hasArgument(.dumpAst), but never actually passed through or used, so maybe that code can be removed:

.indexIgnoreStdlib
.indexSystemModules

[cltnschlosser]

Okay, actually I have to do a bit more digging here. C++ driver claims these are unsupported, so swift-driver probably should too <unknown>:0: error: unknown argument: '-emit-pch'.

This happens for all these parameters. First I'm going to find where this check happens in the c++ driver, then I'm going to look at the unit tests we have that are using these parameters.

[cltnschlosser]

Okay, I've updated the driver to emit an unknown argument error for .noDriver args (this matches c++ driver). Also made a few changes where the swift-driver was expecting these arguments. Including unit tests that used them.

Re-requesting reviews since the diff has changed quite a bit.

Also I need someone to kick off CI :)

[DougGregor]

@swift-ci please test

[owenv]

LGTM! I'm surprised this many options were affected, the C++ driver definitely has some dead code that's checking for these...