LGTM. Sorry I missed this before, I'll commit in a moment with a comment tweak.

Looks good to me.

This is dual licensed and fine per current policy.

• Use "than" where meant.

• Link to the ideas list.

LGTM.

This seems fine, but I'm not a VFS expert by any means. I think the commit message should indicate why this is being done and what the goal is.

Obvious change verified by eyeball.

In D19261#415435, @brooks wrote:

In D19261#415424, @emaste wrote:

In D19261#415418, @drysdale_google.com wrote:

However, I'm merely a kibitzer -- maybe there's a policy/tool constraint that requires reproducing the entire repo?

We have examples of both approaches - importing stripped-down copies of the repo, or the entire, unmodified repo. IMO if there's a only small fraction of upstream that we don't need and could remove it's not worth it, but if we need less than half (say) then it makes sense to me to strip it down.

Right now Phab says we'd bring in 633 files, while AFAICT we really only need around 27 .{c,cc,h} files in the top level capsicum-test directory.

Since google test is GPL it would be nice if it wasn't imported. (EDIT: imported here, specifically, the general import is fine.)

In D19261#415424, @emaste wrote:

In D19261#415418, @drysdale_google.com wrote:

However, I'm merely a kibitzer -- maybe there's a policy/tool constraint that requires reproducing the entire repo?

We have examples of both approaches - importing stripped-down copies of the repo, or the entire, unmodified repo. IMO if there's a only small fraction of upstream that we don't need and could remove it's not worth it, but if we need less than half (say) then it makes sense to me to strip it down.

Right now Phab says we'd bring in 633 files, while AFAICT we really only need around 27 .{c,cc,h} files in the top level capsicum-test directory.

Bring in Jilles to comment on the idea of adding an environmental variable to enable pipefail.

Looks good to me.

FreeBSD's sh grew pipefail support in rS344502 so this might be OBE.

Remove core, the herald rule was over broad.

This looks right to me.

Most files lack required license and copyright headers. The few that do have them indicate this is GPL licensed which is generally unacceptable for new drivers. Ideally this driver should be licensed under the 2-line BSD or dual licensed under it and the GPL if that is necessary. The complete list of commonly acceptable licenses can be found at https://www.freebsd.org/internal/software-license.html

• GC an extra word.

• Reword to make clear that Apache approval is per-component.

Sorry, I had changes in progress when this came in and then life got busy. This one slipped out of my buffer. I'll commit shortly.

I'd be tempted to bump to 192MB since the vast majority of new installs are amd64 per @imp's status.

• Fix typo.

• Rebase
• Reduce the scope of these guidelines to cover only head.

I'm fine with this as is or with the _ENABLE removed.

• Remove obsolete and always true assert.
• Remove pointless close(). fclose() closes the descriptor.

• Make sure there's at least space for one digit of a unit number.

In D18622#408327, @emaste wrote:

@imp, @brooks you seem to be possible curators of contrib/; does google/googletest/ seem like the right location? Seems sensible to me.

LGTM

LGTM

This was committed and should have auto-closed. Commandeer so I can close it.

In D18936#404680, @kib wrote:

Does the removal mean that we are unable to build some old releases in chroot ?

A search of dports for MDIOCLIST shows only on result (in a duplicate header) so I don't think an exp-run is required here https://grok.dragonflybsd.org/search?q=MDIOCLIST&defs=&refs=&path=&hist=&type=