build: reconsider use of -s and -w

[maddyblue]

See:

https://twitter.com/rakyll/status/846860908353634308
https://twitter.com/davecheney/status/846844080227213312
golang/go#19753

[bdarnell]

Note that while we've only recently started using -s and -w, we had been using strip in our beta release binaries for a long time before that (see #14224), with no known issues. In the absence of more specific concerns with stripped binaries, I'm fine with continuing to do so.

[tamird]

That twitter conversation is a marvel of FUD.

[bdarnell]

It turns out that -ldflags -s is also a workaround for golang/go#19741

[maddyblue]

While it may be FUD, those are 2 pretty reliable sources. Dave's argument is that since it isn't tested it's not guaranteed to work. JBD doesn't have an argument, but is on the Go team, so worth considering. I don't think we need to change anything since we've been doing this (-s/-w or strip) for a while now without any visible problems, but we should keep track of that Go issue to see where it goes.

[bdarnell]

As a counterpoint, there's this message from Russ Cox:

To answer the original question, I don't think we have to do anything to "support" this, any more than we "support" readelf or objdump or anything else. We don't intentionally do anything that would make stripping a binary not OK, and strip has worked for the past five years or so. What didn't work was using -s with the linker (that is very different from invoking strip), and even that has worked for the past five years or so.

I don't believe a test in all.bash is needed.

[tamird]

golang/go@94c95d3