added stubs for tcgetattr and tcsetattr, minor node tty fixes

[inolen]

...

[kripken]

Any chance for a test here? :)

[inolen]

I'll write up some tests later today, sure.

[inolen]

Writing the tests for the tty ended up being more work than I'd planned. I decided to make test_stdin work for node. It wasn't that node's stdin read support was fundamentally broken, it's just entirely asynchronous, and the test was updated to treat it as such.

A few minor fixes had to be made to the tty library in order to support reading from text files that were piped to stdin. Currently, the FS startup code doesn't distinguish if stdin is a tty device or not (it always assumes it's a tty device), so the tty library now forces process.stdin to be utf8 encoded. This makes reading from text files work, but really there should be some refactoring to the standard stream initialization such that the TTY device isn't being used when stdin represents a file. Also, there was a really nasty issue where if you read only once from stdin the process would hang in node, I left a short IRC transcript from node's streams author in there explaining the "fix."

Finally, I hit a few snags in libc calls:

• clearerr should reset both eof and error indicators
• fgetc was incorrectly setting the eof indicator. in cases where fread had errored with EAGAIN it was setting eof. I removed the set entirely, as there is no need for fgetc to even worry about it, fread will set the correct value in any case.

[kripken]

Does test_stdin still work on spidermonkey? Let's loop on spidermonkey and node to test both.

[inolen]

Err, I went to try and run it in SM but got errors about setTimeout not being defined:

anthonypesch@lembp:~/projects/emscripten$ js
js> setTimeout
typein:1: ReferenceError: setTimeout is not defined

If this is true (does SM shell not implement an event loop?) I guess I can update the test to manually call main_loop once before hitting emscripten_set_main_loop to be compatible in SM.

[inolen]

Updated the test to work in SM shell, let me know if that comment is incorrect.

Do you want the test runner code to manually iterate over both engines?

[kripken]

Yeah, just loop over [NODE_JS, SPIDERMONKEY_ENGINE].

[kripken]

Yes, no async stuff in sm shell.

[kripken]

Needs a rebase, says it can't be merged.

[inolen]

Rebased, test_pthread_specific was added at the exact same line messing up the auto-merge I suppose.

[kripken]

test_stdin does not pass for me in node, i had to disable it.

I get error exit(1) called.

[inolen]

K, I'll take a look in my VM today. The code in the TTY modules feels kind of bad, but I'd like to get it resolved in node.

[inolen]

While this fixed a few issues, it also created a new issue where Python is exiting with EAGAIN when piping the output during the test runner. You'll notice it when a test fails and all you see is:

IOError: [Errno 35] Resource temporarily unavailable

For whatever reason, by simply referencing process.stdin, e.g.:

process.stdin;

node does something such that reading from its stdout is returning EAGAIN instead of blocking.

Pushed a fix disabling TTY::init and TTY::quit until I can come up with a proper solution as reading the test runner is a bit more important than supporting reading text files through stdin in node:
9451cad

[inolen]

Posted an issue in the node issue queue regarding this:

nodejs/node-v0.x-archive#6175