Re: Linux Downloads page change - Mailing list pgsql-www

From Dave Page
Subject Re: Linux Downloads page change
Date
Msg-id CA+OCxoz7-Guj3861feYOTD59RNw06-5neMmSUZD+vvvvm3NvQg@mail.gmail.com
Whole thread Raw
In response to Re: Linux Downloads page change  (Simon Riggs <simon@2ndQuadrant.com>)
List pgsql-www
On Mon, Jul 9, 2012 at 3:50 PM, Simon Riggs <simon@2ndquadrant.com> wrote:
> On 9 July 2012 13:05, Dave Page <dpage@pgadmin.org> wrote:
>
>> Right - that's more or less what's been discussed and agreed. The
>> issue with the installers that Magnus raised, is that at present I
>> manually push the canonical GIT repo to git.postgresql.org, and often
>> forget to do it until reminded. That was raised in response to my
>> comment that the OpenSCG build scripts are not currently public at all
>> as far as I could see, and should be if their work is to be listed on
>> postgresql.org's primary downloads page.
>
> It's not more or less. What you have said is not the same thing as I
> have requested.
>
> If it was done as I suggest, when you forget a step in the process
> then the process would fail.
>
> If you build from the public repo then you simply can't forget.

The security issue you quote is precisely why we built from the
canonical source, and not a secondary mirror.

You also wouldn't see a failure as you suggest - you'd probably see a
successful build that you later discover is missing recent bug fixes.

>>> Unverifiable binaries are a quality and security risk to the project.
>>
>> In theory. In practice it seems unlikely anyone would ever take the
>> time and energy to build them themselves and actually verify them -
>> the effort to do so would be huge (for example, assembling the 9.2
>> build machine for the installers and building all the necessary
>> dependencies for all the supported platforms etc. has so far taken a
>> number of man weeks). To verify the binaries we put out, someone would
>> have to build an exact mirror of that environment. That's not to say
>> it shouldn't be possible of course. In fact, it wouldn't even be
>> possible, as we digitally sign some of the executables to appease
>> Windows, and we obviously cannot share that certificate.
>
> I know multiple users (aside from 2ndQuadrant) that re-build their own
> binaries as a safety barrier in their release process, so I don't
> believe the effort level is that high, nor do I believe people won't
> do it. I take your point that it is maybe only 1% of people, but those
> are the ones that report all the bugs.

Well if you believe it's that easy, then I'd suggest you try for
yourself. Building the installers is *not* trivial, and building the
installers with an identical dependency tree to verify everything
we've built is a huge undertaking - and as I mentioned, not actually
possible on Windows because you would have no way to sign the binaries
you create with our certificate.

Note again though that we're talking *installers* here, and not RPMs
or other types of packages. The installers are *very* different from
other packages because we have to build so many of the dependencies
ourselves to ensure they'll run successfully on all the supported
platforms.

> The most important thing is that people can see the ingredients before
> they eat the food.

You're welcome to see the code - it's on git.postgresql.org. But that
doesn't mean it would be easy to build a bit-level verifiable copy of
our binaries.

-- 
Dave Page
Blog: http://pgsnake.blogspot.com
Twitter: @pgsnake

EnterpriseDB UK: http://www.enterprisedb.com
The Enterprise PostgreSQL Company


pgsql-www by date:

Previous
From: Simon Riggs
Date:
Subject: Re: Linux Downloads page change
Next
From: Alvaro Herrera
Date:
Subject: Re: 9.2beta web issues