std/cmdline.commandLineParams NimScript fix

[ZoomRmc]

cmdline.commandLineParams() now returns only the script arguments, skipping compiler arguments.

Closes #23554

The logic is extracted from parseopt.initOptParser and is now duplicated. Though there it triggers only when there are no arguments provided so double reads are an edge case:

Read commandLineParams() from a NimScript and get none -> pass them explicitly to initOptParser -> Whole argv reread again and stripped.

Should be fixed if #25571 or an alternative is merged.

[Araq]

The compiler as it evaluates paramCount and paramStr at compile-time should do these fixups.

[ZoomRmc]

Well, maybe it should but it doesn't. These procs are builtins for NimScript, as far as I see. Not sure what you want.

This code is not new, it's just moved copied from parseopt to a more fitting place.

[Araq]

I want you to fix the root cause which is in compiler/scriptconfig.nim, lines:

  cbconf paramStr:
    setResult(a, os.paramStr(int a.getInt 0))
  cbconf paramCount:
    setResult(a, os.paramCount())

[ZoomRmc]

First of all, I wouldn't know how to, I found the source but I'm not very confident at that level.

Secondly, wouldn't this break scripts inspecting the command line launched implicitly from config.nims on compilation?

# in config.nims:
for i in 1..paramCount():
  if paramStr(i) == "-foo":
    raise newException(Defect, "Build with 'foo' is not supported")

[ZoomRmc]

So, continuing the discussion above.

As we have an additional case of implicitly run scripts, I think it would be beneficial to leave the paramCount/paramStr "dumb", being able to access the whole command line. These names bring back Pascal memories so feel more lower-level.

The commandLineParams could be considered a convenience wrapper then. However, if the script is launched implicitly, we don't have the script name in the arguments so we strip everything and get an empty seq. Do we want to return all the arguments or none in this case? Probably the first.

[ZoomRmc]

@Araq I'd like to proceed with this, but I think my concerns above are legit.

To repeat myself, two things need clarifications:

1. Do we really want changing paramStr/paramCount behaviour?
2. What do we want commandLineParams to return if invoked from implicitly launched scripts (config.nims) where no script filename is present in argv?

[ZoomRmc]

Rebased on tip, tests added.

I went with "nothing" as the answer to my second question in the previous post. It's documented and you still have the choice of what to use. Being smarter requires the compiler to expose how the script was invoked in some way.

I think this should be good to go.