build: add long paths awareness on windows

[StefanStojanovic]

On windows when using fs.realpath.native, fs.realpathSync.native and fs.promises.realpath on long path, ENOENT is thrown. On the other hand, long paths work well with fs.realpath and fs.realpathSync. On Linux, all of the mentioned realpath versions work as expected.

The problem with native realpath implementation on windows was, that by default applications aren't long path aware, unless specified otherwise by their manifest file, which was the case with node. The other way to enable working with long paths is to add a \\?\ prefix to path prior to using it in windows API functions. This approach is used by all fs functions working with paths (they add prefix by calling pathModule.toNamespacedPath(path)), but this is not done in fs.realpath.native and fs.realpathSync.native.

Although fix could have been made in fs.js, a change to node manifest file was chosen as a more comprehensive and broader solution to the long paths on windows problem in general.

An existing windows long path test is improved to cover the issue that is fixed by these changes.

Fixes: #39721

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/46426/

[Flarna]

@StefanStojanovic It seems the test failure is related to your change. Please take a look.

[StefanStojanovic]

@StefanStojanovic It seems the test failure is related to your change. Please take a look.

Yes, tests failed because of my change. The problem with it is that manifest file long path awareness relies on registry key HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\FileSystem\LongPathsEnabled existing and being set to 1 plus it only works on Windows 10 version 1607 and later. So, since this solution isn't guaranteed to work in all cases, I made another PR focusing the fs.reaplath.native issue specifically: #44536

[lpinca]

Does it make sense to land both this and #44536?

[lpinca]

Does it make sense to land both this and #44536?

I wonder if it is better to use a single commit to facilitate backporting without breaking tests.

[StefanStojanovic]

Does it make sense to land both this and #44536?

My idea was to just land #44536. Since manifest file fix isn't guaranteed to work, I'm thinking this PR should probably be closed, just wanted to keep it open for reference until the other one lands.

[lpinca]

Ok.

[lpinca]

Superseded by #44536.

[Flarna]

@StefanStojanovic I guess we can close this one as #44536 landed?

[StefanStojanovic]

@StefanStojanovic I guess we can close this one as #44536 landed?

Yes