fs: fix cpSync crash on utf characters

[jazelly]

Fixes: #54476
Refs: #54653 (comment)

[codecov]

Codecov Report

Attention: Patch coverage is 86.04651% with 6 lines in your changes missing coverage. Please review.

Project coverage is 88.04%. Comparing base (59c7c55) to head (32faca8).
Report is 48 commits behind head on main.

Files with missing lines	Patch %	Lines
src/node_file.cc	85.00%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #54653      +/-   ##
==========================================
+ Coverage   88.03%   88.04%   +0.01%     
==========================================
  Files         652      652              
  Lines      183763   183779      +16     
  Branches    35860    35865       +5     
==========================================
+ Hits       161774   161817      +43     
+ Misses      15237    15228       -9     
+ Partials     6752     6734      -18     

Files with missing lines	Coverage Δ
src/util.h	90.16% <100.00%> (+0.24%)	⬆️
src/node_file.cc	77.04% <85.00%> (-0.06%)	⬇️

... and 37 files with indirect coverage changes

[aduh95]

It looks like it's consistently failing to build on our macOS GHA CI

[jakecastelli]

Possibly related to this

ld: write() failed errno=28

which indicates no space left on the device, I assume that is why it fails

[jazelly]

With the requested CI env, I can finally reproduce the segfaults.

The original reported issue
Chinese are UTF-8 encoded and each char is 3-byte encoded. 5 characters are 15 bytes in the reported case. Windows std::filesystem::path treats it as UTF-16 encoded, which is 2 bytes per char, and hence the segfault happened when constructing the std::filesystem::path for src or dest.

Simply constructing from u8string still has segfault (this was the one I couldn't reproduce locally, very strange)
The segfault comes from the later call of std::filesystem::path.string() on the path constructed from u8string. I don't particularly know why? But using u8string for all the error messages fixed it. Thanks @jakecastelli for requesting the CI environment. (sorry for dragging this too long).

cc @joyeecheung @anonrig

[joyeecheung]

The segfault comes from the later call of std::filesystem::path.string() on the path constructed from u8string. I don't particularly know why? But using u8string for all the error messages fixed it.

This is because the error throwing functions convert the messages into JS strings - technically, it is expecting Latin1

node/src/node_errors.h

Line 121 in eab97b5

OneByteString(isolate, message.c_str(), message.length()); \

which is also incorrect, but Latin1 has an overlapping set with UTF-8, and if you pass UTF8 data as Latin1 you are likely to see gibberish for non-ASCII characters but it won't crash, whereas UTF16 has fixed 2 char widths, so it would fare worse if you pass it as Latin1.

[joyeecheung]

LGTM - I opened a separate PR to fix the decoding in error throwing utils #55024

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/62589/ 💛

[jasnell]

Landed in 77ca5ca