net/http: test spams stderr

[bradfitz]

Running go test on a package should be quiet and not spam to stderr.

But:

$ go test -short
2018/04/12 17:09:17 http: WriteHeader called with X-Content-Type-Options:nosniff but no Content-Type
2018/04/12 17:09:17 http: TLS handshake error from 127.0.0.1:55706: remote error: tls: bad certificate
PASS
ok      net/http        4.034s

The first log line was added in 1a677e0 (/cc mikesamuel).

I haven't looked into the second.

Help fixing these would be great. Thanks!

[odeke-em]

Thank you for this report @bradfitz

For further investigation, I'll kindly page:

1. /cc @mikesamuel -- @bradfitz you forgot the @ :)
2. Looks like a regression from c529141 /cc @mikedanese @FiloSottile @agl

[mikesamuel]

Sorry. The first one's my fault. I'll look into it. The second looks unlikely to be caused by what I did to cause the first.

[odeke-em]

@mikesamuel no worries :) In deed, the second one was caused by a different commit

Looks like a regression from c529141 /cc @mikedanese @FiloSottile @agl

[gopherbot]

Change https://golang.org/cl/106756 mentions this issue: net/http: log output from serve_test reached stderr

[mikedanese]

Haven't looked deeply into this yet. @odeke-em can you give me a hint on what lead you to c529141?

[mikedanese]

I don't think c529141 is related. More likely 2638001.

[odeke-em]

Sorry for the late reply @mikedanese, I've been travelling and was away from my computer for a bit. I went through the history https://github.com/golang/go/commits/master/src/crypto and commits before and after yours after some bisection using some hacked up bash script to do binary searches, recompile the Go tree and then find regressions(but it isn't yet complete so not yet releasing it ;))

@mikedanese you can try to confirm it yourself, the commit before yours 8a15192 didn't produce that result but yours c529141 did and so did the commit after yours 0b37f05

[mikedanese]

From my previous comment, I've ruled out c529141. You can confirm this by reverting that change, running tests, and still seeing the error output. The test added in 2638001 is creating the output, and has been creating the output since it was merged.

[odeke-em]

@mikedanese thank you for the thorough investigation, I was just following changes in crypto, my apologies for the spurious git blame.

In that case /cc @pquerna and @FiloSottile.

[gopherbot]

Change https://golang.org/cl/107817 mentions this issue: net/http: remove stderr log spam in test