Opening as draft; only the last four commits are new

@tianon ptal (probably easiest to review commits individually)

I'm a little confused about still having all the strings around ("This function still receives the 'raw' string, as it's used to include the "raw" spec for inclusion in error messages.") -- if it's just used for error messages, regexp.Regexp has a String method that claims it returns the original expression back again, and that seems like it'd be enough?

@tianon had to read back what I wrote there, but that's about this function:

func (p *windowsParser) parseMount(arr []string, raw, volumeDriver string, destRegex *regexp.Regexp, convertTargetToBackslash bool, additionalValidators ...mountValidator) (*MountPoint, error) {

It receives both

• the arr []string (volume-spec split into its components)
• the raw string (volume-spec before it was split)

Last one is used because for some errors, it uses the original (pre-split) volume-spec to generate an error;

And (of course) I thought: that's easy, I just remove the raw argument, and generate the error in ParseMountRaw() (the exported function);

m, err := p.parseMount(arr, raw, volumeDriver, mountDestinationRegexp, true, windowsSpecificValidators)
if err != nil {
    return nil, errInvalidSpec(raw)
}

But then found that

• there's various errors it generates; errInvalidSpec, errInvalidMode, fmt.Errorf("%v: %v", errInvalidSpec(raw), err)
• for the above, I was in doubt wheter or not wrapping errInvalidMode into a errInvalidSpec would be ok (perhaps it is)
• for the errInvalidSpec errors, there is no error message to wrap (perhaps it could return a ErrInvalidMode (invalid volume specification), which we then format.

@tianon perhaps something this? d2eb3cd

Ah, I guess I was hoping there'd be a cleaner way to avoid keeping all the string variables too (like rxDestination) so we wouldn't accidentally use one of them instead of using the MustCompile'd versions, but I see now that they're mostly reused for multiple underlying expressions, so that's not super realistic. 😞

So I guess I'm more neutral on whether #42667 is worth solving by itself, but I think doing MustCompile in the package instead of in a function is probably better (since the error will be noticed sooner), and it looks like you've done other cleanups here too that look useful IMO.

Yeah, doing it once is probably still a good thing to do https://twitter.com/kolyshkin/status/1419025281780654081?s=20, that is, if it's ok to allocate a couple of kilobytes (saw @kolyshkin tweet about this recently 😂)

@tianon were there things left to address, or all good? 🤗