`stdin.read` (and `stdin.readSync`) corrupt non-ASCII input on Windows

[lionel-rowe]

stdin.read (and stdin.readSync) corrupt non-ASCII input on Windows.

To reproduce:

const c = new Uint8Array(6);
Deno.stdin.read(c).then(() => console.log(c));

Then, enter a non-ASCII character. The resulting bytes will be corrupted on Windows.

Examples, with trailing LF/CRLF/null bytes truncated:

Input	Expected	Actual	Decoded as UTF-8
ÿ	[195, 191]	[152]	"�" (Invalid char)
Ā	[196, 128]	[65]	"A"
ā	[196, 129]	[97]	"a"
啊	[229, 149, 138]	[63]	"?"
🦄	[240, 159, 166, 132]	[63, 63]	"??"

Expected results are the UTF-8 bytes. Results on Linux are as expected.

[njhanley]

This stems from Deno reading directly from console input as a file, which uses the console's current code page rather than Unicode (see High-Level Console I/O). In the example, 'ÿ' maps to 152 in code page 437 (OEM-US).

Instead ReadConsoleW should be used, as in Rust's std::io::Stdin.

@dsherret Is there a reason Deno doesn't use Rust/Tokio's stdio implementation? If not, would a PR be welcome?

[dsherret]

Yes, a PR would be welcome. I believe it should use std::io::stdin here when StdFileResourceKind::Stdin:

https://github.com/denoland/deno/blob/3cd7abf73fa104526508984daef54bbb8e120310/ext/io/lib.rs#L398-L400

Similar to how it does this for write:

https://github.com/denoland/deno/blob/3cd7abf73fa104526508984daef54bbb8e120310/ext/io/lib.rs#L377-L383

[Mqxx]

Hey, I just stumbled across this issue having the same problem. Characters like ÄÖÜ (non ASCII) are corrupted. Any update on when this gets fixed?

Thanks