Large preprocessor outputs hit the maxBuffer limit

[freejosh]

When using one main file that imports all my CSS with a preprocessor I was hitting the maxBuffer limit. I've changed exec to spawn so that the output gets built from the stream, without a limit.

Also fixed a typo in the regex that checks for Sass partials.

[marrs]

Thanks for the request. Can you provide a regression test for child_process.spawn and resubmit?

[freejosh]

I'm not too sure how to do that, do I make a file like test/urlRelative.js? If that's what you mean, would you say it's fine to skip passing in an actual preprocessor command, and just run cat to read the contents of a large CSS file?

[marrs]

Yes, exactly. The urlRelative test is written using Mocha and Chai. The others are written in NodeUnit. You can use whichever you find more comfortable.

If reading the contents of a large file directly still reproduces the error that you fixed then I would say it's preferable to skip the pre-processor command. Ideally we want the fewest/simplest conditions to reproduce the error.

[freejosh]

Let me know if adding the tests here was fine, or if they should be in a separate pull request, to test with unfixed code first.

[marrs]

Here is fine. I trust that you watched the test fail and then made it pass. :)

Although, now that I understand the nature of the bug, I'm wondering if it's worth including this test. I'm not sure I want to add a 200M stub to the repo just to reproduce, but without it the test is at best useless.

Let's merge the branch without the test. I think a better idea for the future is to hook a linter into the test suite to check the code for child_process.exec and warn the user against using it.

[freejosh]

Sounds good. Yea the warning should be something like:

exec, by default, only returns 200k of data. Use spawn if you're expecting larger return data.

[marrs]

Thanks again for the patch. It'll be in the next maintenance release