Re: [boost] [1.45] Typeof fixed, point release? (was: Boost 1.45 compile error)

Subject: Re: [boost] [1.45] Typeof fixed, point release? (was: Boost 1.45 compile error)
From: Vladimir Prus (vladimir_at_[hidden])
Date: 2010-11-27 01:39:02

Beman Dawes wrote:

> On Sun, Nov 21, 2010 at 5:34 PM, David Abrahams <dave_at_[hidden]> wrote:
>> At Sun, 21 Nov 2010 14:27:46 -0800,
>> Robert Ramey wrote:
>>>
>>> Eric Niebler wrote:
>>> > I have just fixed the typeof problem on the release branch. (I've also
>>> > verified that it makes the common_type regression go away without
>>> > causing additional failures for either the typeof or type_traits
>>> > libraries.)
>>> >
>>> > I know it's not our policy to issue point releases, but totally
>>> > breaking
>>> > a major header (boost/type_traits.hpp) for a primary test compiler
>>> > (msvc-7.1) seems like a good reason to make an exception. Does anybody
>>> > feel otherwise?
>>>
>>> Since msvc-7.1 doesn't appear in the test matrix for the release version
>>> I don't see how it can be considered a primary test compiler.
>>
>> +1
>>
>> Now, maybe the answer is "we messed up by not noticing that our tester
>> upgraded to vc-8" or something.  But either way, the release managers
>> have to decide if this is important enough to issue a point release.
>> VC-7.1 _is_ getting a bit long in the tooth at this point...
> 
> The current situation is a total mess in that the release managers
> have no practical control over who tests which branch with what
> compiler. We need a pool of available platforms, and a mechanism
> controlled by the release managers to parcel out tests to the
> available platforms.The only tests that mechanism will run are release
> platform tests on the release branch. Then there needs to be a
> different interface that a developer uses for test-on-demand of
> his/her Git repository. Then the GIT push or pull mechanisms can be
> used to move code into the main Boost release repository only when it
> is already passing tests. The distinction between trunk and release
> goes away.

I don't think so. You're replacing release and trunk branches in a same
repository, with one release branch, and N work-in-progress branches in
different repositories, which makes the whole situation less manageable,
not more.

(For example, you might want to drop into boost IRC channel and count
the number of people who complain about 10 git clones of Boost, in
various degree of deadness).

- Volodya