John Sonnenschein wrote:
> 
> On 13-Nov-08, at 2:23 PM, Shawn Walker wrote:
> 
>> John Sonnenschein wrote:
>>> On 13-Nov-08, at 1:59 PM, Nicolas Williams wrote:
>>>> On Thu, Nov 13, 2008 at 06:20:24PM +0100, Guido Berhoerster wrote:
>>>>> * Shawn Walker <[EMAIL PROTECTED]> [2008-11-13 17:23]:
>>>>>> John Sonnenschein wrote:
>>>>>>> I'd just like to throw my thoughts in to the ring for this, but the
>>>>>> genunix page lists "Binary only packages allowed" as a goal..
>>>>>>>
>>>>>>> That is, in my opinion, a /TERRIBLE/ idea, and likely to get people
>>>>>> in more trouble and reflect worse on (open)solaris than just 
>>>>>> simply not
>>>>>> having the packages at all. Any sort of poor software, either through
>>>>>> malice (trojans) or incompetence (running being heavily dependent 
>>>>>> on the
>>>>>> builder's specific system setup) can be sneaked in with absolutely no
>>>>>> oversight.
>>>>>>
>>>>>> That's assuming that the packages don't:
>>>>>>
>>>>>> 1) receive any vetting at all
>>>>>
>>>>> So how does the reviewer make sure (with reasonable effort) that
>>>>> the submitter has not injected malicious code in the binary package
>>>>> he submitted?
>>>>
>>>> The reviewer can't really know that even if source is provided, not as
>>>> long as the reviewer accepts object code built by the submitter.
>>>>
>>>> I think you may want to argue that submitters should submit spec files
>>>> for building things and let trusted providers build the actual 
>>>> packages.
>>>>
>>>> In a way we'll be doing just that.  First, we'll be our own submitters
>>>> of spec files.  Second, we'll review and use spec files that exist
>>>> already or that others contribute.  The main caveat is that if a spec
>>>> file includes patching of third party FOSS source then we'll need to
>>>> complete an OSR, whereas if we don't modify FOSS source then the 
>>>> process
>>>> is lighter-weight.
>>>>
>>>> That doesn't mean that we'll only accept spec files.  We currently
>>>> intend to accept binary-only pkgs into the /contrib repo, and we intend
>>>> to tag them accordingly.
>>> I still wholeheartedly maintain that binary-only packages ought to 
>>> be  an exception to the rule and process should try to avoid it as 
>>> much as possible.
>>> That is, you ought to have a decent reason to be distributing 
>>> closed-only packages, and people who are trusted ( the people in 
>>> charge of contrib/ ) should be able to vouch for your trustworthiness.
>>> Letting some random user upload an unknown and untrusted binary is a 
>>> recipe for disaster
>>
>> ...and I want to be clear that I'm not arguing against that.  That is 
>> a perfectly reasonable set of guidelines for many software packages.
>>
>> However, one size does not fit all, and that should also be recognized.
>>
>> Not every package should have to have a spec file; trusted members 
>> should be able to deliver to a staging server for review directly.
> 
> Exactly!
> 
> I'm not arguing that the process must fit everything always, just that 
> the process ought to streamline the stuff that does fit in to the mold ( 
> if you can build a .spec file, it shouldn't require human intervention 
> ), but if you are a special case ( weird build instructions, binary only 
> ) a little bit more oversight is necessary

Then Huzzah!; we're in agreement.

-- 
Shawn Walker
_______________________________________________
indiana-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/indiana-discuss

Reply via email to