src: make SIGPROF message a real warning

[cjihrig]

This commit replaces a fprintf() with a call to ProcessEmitWarning().

R= @Fishrock123

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• commit message follows commit guidelines

Affected core subsystem(s)

src

[addaleax]

It would be good to have a test if that’s possible (it might be?) but LGTM either way

[jasnell]

Agree with @addaleax that a test would be good if possible but otherwise LGTM

[Fishrock123]

Test and CI, but otherwise LGTM

[cjihrig]

So it looks like the inspector is always started by default, making the surrounding env->inspector_agent()->IsStarted() check always true. That means that the warning will always be emitted if process.on('SIGPROF', ...) is used in the application. That seems wrong to me.

[Fishrock123]

@cjihrig That sounds incorrect to me too. That is why I was wondering in the issue why the event emitted even though we were not using the inspector...

[sam-github]

We always attempt to start the inspector, but it doesn't necessarily succeed. I follow

node/src/node.cc

Line 3986 in 6bcf65d

debugger_running = v8_platform.StartInspector(env, path, debug_options);

down the bottom I get lost in the weeds, but it does appear that some low-level error conditions can cause the inspector to fail to start.

Also, building without the inspector cause it not to start :-)

Since SIGPROF is for the profiling timer (see setitimer(2)), maybe always warning about it is a good idea? On the one hand, it removes a signal from the useable set, but on the other, it sucks to write code that later fails to work when its run under the debugger.

[cjihrig]

Right, we always attempt to start the inspector unless Node is built without it. If others are ok with this behavior, I can add a simple regression test.

[cjihrig]

Test added. CI: https://ci.nodejs.org/job/node-test-pull-request/7993/