Support NO_PROXY and no_proxy

[sgoff0]

While mentioned in a needle Pull Request comment I believe lack of NO_PROXY support should be called out as its own issue.

I noticed this behavior while using Pact, a contract testing library, as it previously used request but replaced it with needle. With the move to needle http requests started failing behind our corporate proxy as needle does not respect the no_proxy environment variable we have set for localhost. Specifically in our case http requests via needle to localhost (something Pact does when running a contract test) were going through our system proxy which ended up returning a 302 to a proxy blocked website.

This issue is resolved if config.proxy were to be set to null prior to the needle.js proxy check when a URI is in the NO_PROXY environment variable

needle/lib/needle.js

Lines 302 to 315 in 4cb2ef6

	// if proxy is present, set auth header from either url or proxy_user option.
	if (config.proxy) {
	if (config.proxy.indexOf('http') === -1)
	config.proxy = 'http://' + config.proxy;
	if (config.proxy.indexOf('@') !== -1) {
	var proxy = (url.parse(config.proxy).auth || '').split(':');
	options.proxy_user = proxy[0];
	options.proxy_pass = proxy[1];
	}
	if (options.proxy_user)
	config.headers['proxy-authorization'] = auth.basic(options.proxy_user, options.proxy_pass);
	}

As an example in my use case manually adding config.proxy = null right before this check makes the network call behave as expected. This obviously is not the right fix just demonstrates that manually setting null here when desired seems to solve the problem. As an FYI the now deprecated request package already has logic to properly handle NO_PROXY and no_proxy, some of this could likely be leveraged here to add support for NO_PROXY

[tomas]

Thanks for the heads up. I'll add NO_PROXY to the PR and merge it soon.

[nejch]

Thanks for the work on that PR @tomas. I think this issue can remain open for now to make sure needle can address the syntax usually used in no_proxy described above. For example:

https://github.com/w3c/libwww/blob/8678b3dcb4191065ca39caea54bb1beba809a617/Library/src/HTAccess.c#L234-L239

Usually no_proxy will be used with domain wildcards, e.g. no_proxy=.company.com,.company.io, which should cover all ports and all subdomains for those 2 top-level domains. There is also some discussion in curl/curl#1208 that might be relevant.