Comment 1:

Labels changed: added priority-soon, go1.3maybe, removed priority-triage.

Comment 2:

Owner changed to @agl.

Comment 3:

Out of luck?
In https://github.com/bradfitz/camlistore/blob/master/pkg/client/android/android.go I
use x509.NewCertPool() to make my own *tls.Config.

Comment 4:

Okay, maybe not out of luck. But it'd be nice to make it easier for cert verification to
work without having to do manual wiring.

Comment 5:

Unclear we want to give more control over this than what's already there.

Comment 6:

It's not really about more control, but just making it easier to use the control that is
already present. e.g. if I'm on a system that puts the SSL system roots in some weird
place, what do I do? When *I* read through the package I didn't find the
x509.NewCertPool approach that Brad used, so at the least I think the docs could make it
clearer (maybe an example?).

Comment 7:

The question is whether one package you import should be able to override
the system x509 location for others. Normally we trust packages to behave
but this is a fairly serious thing to be able to change.
In package time we use $ZONEINFO. Perhaps a similar environment variable
makes more sense than having each program change it, but even then I am a
little worried.
Certainly we should have a good example.

Comment 8:

Maybe from a safety perspective the API exposed can be limited to only *appending* to
the system root list, so it'll only be effective if the system can't find the roots in
the usual places. That reduces the attack vector from being able to subvert any cert
validation to only being able to permit cert validation when it otherwise would be
unable to be validated.
But you're right it needs some careful thought. Are we considering one package being
able to attack another (e.g. a reflection attack)? I guess that's only relevant if
places don't take a *tls.Config but only ever use the default, since you're still able
to load arbitrary roots into a custom tls.Config.

Comment 9:

I'm a little worried about attacks but more worried about inadvertent
breakage, where a package decides to install its own root list as the
"system" list because that's all *it* needs and doing so is easier than
preparing a tls.Config, and then other packages that need the standard list
break.
I also don't think it's so terrible to just edit the source code if you're
compiling the standard library for a system with a non-standard place.

Comment 10:

Labels changed: added release-none, removed go1.3maybe.

Comment 11:

Labels changed: added repo-main.

Comment 12:

Let's make this a documentation bug and get it fixed for 1.3.

Labels changed: added documentation, release-go1.3, removed packagechange, release-none.

Comment 13:

This issue was closed by revision d4d7705.

Status changed to Fixed.