refactor: split path into per-os modules

[jollytoad]

Splits path functions into a module per platform and common fns.
Also applied the pattern to glob.

As discussed in #3553
Also see #3600

[CLAassistant]

All committers have signed the CLA.

[kt3k]

This looks nice

[kt3k]

Maybe web can move normalizeGlob and joinGlobs to _common like isGlob?

[kt3k]

Is this kept because of backward compatibility? Do you think it's good idea to keep this option in the future?

[jollytoad]

This was mainly so that the tests still work, personally I think the os option should be ditched entirely otherwise it's inconsistent with the other path fns, but the tests will need reworking to account for that.

[kt3k]

I wonder if we should also deprecate this path (and also path/posix.ts).

[jollytoad]

It would be useful to be able to deprecate an entire module, so that it's import statement and/or all imported members display with usual deprecated strike-through, but I'm not sure there is a way to do that is there?

[kt3k]

Probably need to alias items and deprecate the aliases.

import * as path from "./windows/mod.ts";
/** @deprecated (will be removed after 0.210.0) Import from `path/windows/mod.ts` instead. */
export const sep = path.sep;
/** @deprecated (will be removed after 0.210.0) Import from `path/windows/mod.ts` instead. */
export const delimiter = path.delimiter;
/** @deprecated (will be removed after 0.210.0) Import from `path/windows/mod.ts` instead. */
export const resolve = path.resolve;
...

[kt3k]

This should be backward compatible from my understanding. Mostly looks good to me.

I wonder if it makes sense to have path/posix/to_namespaced_path.ts as it doesn't have effect in non-window platforms.

[jollytoad]

I wonder if it makes sense to have path/posix/to_namespaced_path.ts as it doesn't have effect in non-window platforms.

I think it makes sense to keep it purely for the sake of API consistency between /path, /path/posix & /path/windows.

[kt3k]

I think it makes sense to keep it purely for the sake of API consistency between /path, /path/posix & /path/windows.

Ok. Makes sense.

[kt3k]

Is there anything remaining with this PR?

[jollytoad]

Maybe web can move normalizeGlob and joinGlobs to _common like isGlob?

I was just looking into this and noticed that there is an inconsistency in the options. The GlobOptions jsdoc says globstar defaults to true, which is the case for globToRegExp, but in normalizeGlob & joinGlob it actually defaults to false!

Also, these fns are pretty simple and rely on platform specific imports (eg. SEP, SEP_PATTERN, normalize, join), which would have to be passed into a common fn, I think it may just be simplier to retain the platform specific implementations of each.

[kt3k]

LGTM. Nice work!