John Plocher wrote:
> James Carlson <carlsonj at workingcode.com> wrote:
>
>> .. Thus "FOSS is special."
>>
>
> I believe FOSS *IS* Special - because doing a good job of integrating
> general cross-platform FOSS into OpenSolaris is actually HARDER than
> integrating something invented by the community specifically for the
> OS itself.
>
I disagree. Its only necessarily harder if you want to have arbitrary
goals like 100% source fidelity to the upstream. (And then there are
cases that are harder almost by design -- such as "hal" or "devicekit",
which integrate tightly with the rest of the OS. But I don't think
we're talking about eliding -- or even reduced -- review for those bits,
are we?)
The larger question is, at what point does it become acceptable to
abdicate architecture (and in some cases quality) considerations just to
remain "compatible" with Linux. (And "compatible" here means different
things to different people. Some people will argue that if their
favorite command isn't installed by default, or has different switches
or different output, that it isn't compatible. Others will be happier
for API compliance. And still others will say its compatible only if
their Linux binaries work on Solaris (which we can do with branded zones.)
As much as some people thing that the current solution is busted, I
disagree. I think we *need* humans to look at cases, and help determine
whether something really is appropriate for inclusion as is, or whether
more work (architectural or otherwise) is needed. I think total
abdication of architecture review for *any* piece of software which we
believe should be "integrated" is *always* wrong.
If you want more of an "if it compiles integrate it" kind of OS, then I
kindly refer you to slackware, who at least historically, have taken
this approach. (Not sure about anymore.) I don't believe that our
biggest value in OpenSolaris is in the ways we are the same as Linux. I
believe are biggest value is in the ways that we are *different*.
(ZFS, Dtrace, Crossbow, just to name a few.) In some cases, we even
improve FOSS software by going beyond the bar and doing more. (Dtrace
entry points for Apache, MySQL, etc? Tighter integration with our
security framework for OpenSSH. Support for our crypto APIs in
OpenSSL. Etc.)
Architecture review for *integration* is IMO *always* appropriate. I
don't think we should give a free ride based on where the stuff came
from. We *can* however elect to say that architecturally compatibility
with other platforms is more important than some other features which we
might have requested if the software were developed exclusively for
Solaris. We *can* also look at a given piece of FOSS to determine that
it needs either more or less scrutiny, based on precedent and history on
other FOSS platforms. (But not that in this case also, at least there
is an active decision to do this. Nothing is "automatic" just because
its FOSS software.)
Remember also, we do *not* have to integrate *every* random bit
o'software. We have a /contrib repository. Users can still supply SVR4
packages. Or tarballs. Or just source code. IMO, its a syntax
error for us to be trying to get every invented FOSS package onto
OpenSolaris. Sure, some bits are very important because of their wide
popularity, and perhaps breaking interfaces for those bits would be
considered unacceptable.
But other bits aren't. How many sites are really going to be upset if
"unison" isn't automatically packaged for them? Or for that matter, GNU
hello? I believe it is perfectly reasonable to allow users to push
their favorite bits into /contrib, skip the ARC review, and allow the
community to support those packages on a caveat emptor basis.
The rest of the stuff, we should continue reviewing, using the same
faculties that we have heretofore been using -- intelligent review by
human engineers. I contend that the existing process works for these cases.
-- Garrett
