Re: [gentoo-portage-dev] [PATCH] egencache --update-changelogs: fix timestamp assumptions (bug 579292)

[Zac Medico]

On 04/08/2016 06:12 PM, Brian Dolbec wrote:
> On Fri, 8 Apr 2016 00:29:40 -0700
> Zac Medico <zmedico@g.o> wrote:
>
>> On 04/07/2016 11:51 PM, Brian Dolbec wrote:
>>> the above looks good, but what about:
>>>
>>>
>>> [19:01] <dwfreed|phone> just use --first-parent
>>> [19:01] <dwfreed|phone> also take into account merge commits
>>> [19:03] <dwfreed|phone> git really complicates ChangeLog generation
>>> in general [19:03] <dwfreed|phone> because your ChangeLog should
>>> reflect when these commits became part of master, but you still
>>> need to perserve their messages [19:04] * zmedico is skeptical
>>> about the linearizability of the timestamps [19:04] <dwfreed|phone>
>>> if you don't look at merge commits for your timestamps of changes,
>>> correct, it is not linear [19:05] <dwfreed|phone> but if you take a
>>> set of commits and determine when they became part of master, it is
>>> linear [19:05] <dol-sen> so: git log --first-parent --format=%ct
>>> -1 . [19:05] <dol-sen> to get the last timestamp of changes to htat
>>> pkg [19:06] <dol-sen> then use that timestamp [19:06] <dol-sen>
>>> lmod = self.grab(['git', self._work_tree, 'log', '--format=%ct',
>>> '-1', '.']) [19:06] <dol-sen> that is the current code [19:06]
>>> <dwfreed|phone> git log -m --first-parent --format=%ct -1 . [19:06]
>>> <dol-sen> so just add the --first-parent option? [19:07]
>>> <dwfreed|phone> you want -m toolmod = self.grab(['git',
>>> self._work_tree, 'log', '--format=%ct', '-1', '.'])
>>>
>>> [19:06] <dwfreed|phone> git log -m --first-parent --format=%ct -1 .
>>> [19:06] <dol-sen> so just add the --first-parent option?
>>> [19:07] <dwfreed|phone> you want -m too
>>>
>>>
>>> Don't we need to add the -m --first-parent  ???
>>>
>>
>> I don't know enough about how those options matter for git log
>> behavior, but I trust that dwfreed has good reasons to recommend
>> them. Maybe we should add them in a separate patch, possibly with
>> some explanation about how they are useful in this context.
>
> They really matter when the tree gets crappy/stale merge commits from
> pull requests which play with the way the tree is considered.  In this
> particular case, it will ignore the branch history and only consider the
> merge commit that connects it to master.  Which is when we want to make
> the changelog entry, not when the (possibly stale) user commit was made.

Bug filed:

https://bugs.gentoo.org/show_bug.cgi?id=579402
-- 
Thanks,
Zac