james-server-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Stefano Bagnara <apa...@bago.org>
Subject Re: [mime4j] what else need to be done (Was: Benchmark of 0.3 vs 0.4)
Date Tue, 05 Aug 2008 16:07:18 GMT
Bernd Fondermann ha scritto:
> On Tue, Aug 5, 2008 at 13:41, Niklas Therning <niklas@trillian.se> wrote:
>> Stefano Bagnara wrote:
>>> Niklas Therning ha scritto:
>>>> Oleg Kalnichevski wrote:
>>>>> Stefano Bagnara wrote:
>>>>>> Robert Burrell Donkin ha scritto:
>>>>>>> what else needs to be done before we can ship?
>>>>>> I'm looking here:
>>>>>> https://issues.apache.org/jira/browse/MIME4J?report=com.atlassian.jira.plugin.system.project:roadmap-panel
>>>>>> I see 4 issues open for the 0.4 release:
>>>>>> MIME4J-57 Add a max limit to header length for parsing.
>>>>>> https://issues.apache.org/jira/browse/MIME4J-57
>>>>>> - this seems critical because it may results in OOM/DoS, but we had
>>>>>> this in past too, so could even be moved to 0.5.
>>>>>> MIME4J-69 Decoding/encoding is not coherent between headers and body
>>>>>> https://issues.apache.org/jira/browse/MIME4J-69
>>>>>> - this probably is too complicate to delay a release and I don't
>>>>>> the evergy to discuss how it should be correctly solved, now. So
if no one
>>>>>> plan to work on it soon, it should be moved to 0.5.
>>>>>> MIME4J-51 Remove cyclic dependencies and provide better organization
>>>>>> the source tree.
>>>>>> https://issues.apache.org/jira/browse/MIME4J-51
>>>>>> - I applied my proposed patch. There are concerns from you and Bernd
>>>>>>  1) remove also the util package.
>>>>>>  2) add a package documentation with examples and "parser" references.
>>>>>> I personally don't care of #1, and if needed I can work on #2 but
>>>>>> without examples, simply adding a package.html with one single sentence:
>>>>>> "the main classes for the pull and SAX parser are in the parser package."
>>>>>> MIME4J-27 [JW#7] Limitations Support
>>>>>> https://issues.apache.org/jira/browse/MIME4J-27
>>>>>> - No answers from Jochen to your question in a month. maybe it should
>>>>>> be moved to 0.5.
>>>>>> Once the 4 issues above have been solved (or postponed) we need a
>>>>>> release manager.
>>>>>> Nothing else I can remember now.
>>>>> Stefano,
>>>>> None of these issues seems severe enough to block the 0.4 release. I
>>>>> would very much appreciate if the 0.4 release could happen rather sooner
>>>>> than later. It would be very unfortunate if we had to release HttpClient
>>>>> 4.0-beta1 without the HttpMime module.
>>>>> Oleg
>>>> I agree with Oleg. IMO these issues can be moved to 0.5.
>>>> /Niklas
>>> Just to make it clear that I agree with you two, too. I simply dumped the
>>> JIRA status and told that I had nothing against moving every open issue to
>>> 0.5. So, if no one object on this we now miss only the volunteer that will
>>> act as the release manager.
>> Ok, great. I have no idea of what it takes to do a release I'm afraid.
>> Someone else?
>>> After your benchmark I applied the package refactoring I proposed in
>>> MIME4J-51, so I'd like you and Oleg to try updating your client code to see
>>> how many changes are needed and if the new structure make sense. This is
>>> something I would avoid changing too often, so it is better to vet it now
>>> before we write it in the stone with a release.
>> I'm afraid I have missed the whole "review packaging" discussion, I hope I'm
>> not too late. :) I'd prefer to have MimeStreamParser and MimeTokenStream in
>> the root package if possible. That would be the most natural thing I think
>> since those are by far the most central classes. I also think ContentHandler
>> and AbstractContentHandler could go in the root package. With these changes
>> the upgrade effort for me would be close to none since what we use is
>> probably only MimeStreamParser, ContentHandler and BodyDescriptor. I
>> understand that this would introduce some cycles but I still think that it
>> would make Mime4j a lot easier to use for the end user. I'd rather have some
>> cyclic dependencies if it makes the whole thing easier to understand.
> +1
> Analysing package dependency statistics and correcting most cyclic
> dependencies is good and very helpful. In my opinion, going as far as
> putting up a zero-cycles-policy makes more user-friendly and pragmatic
> solutions like the proposed one impossible. I am repeating myself by
> saying that exposing the central classes in the root package is by far
> the most user friendly.

Niklas, Bernd, please can you make a concrete list of changes you 
propose (against current trunk or against the tree before my change) if 
this is not one of the 4 solutions I listed in a recent answer to this 
thread so that we can add a 5th solution to this issue and we can cast 
our preferences and see where the consensus build?

I happen to share only the concern that this change require updates for 
old version users. I don't shared the "user-friendly" and "pragmatic" 
concerns. IMHO a parser package is much more user-friendly and pragmatic 
than having 20 classes in the main package even if I ignore the cycles 
issue. So it is matter of opinions, we should simply poll to see where 
the majority is.

Good packaging and no cycles is also very important if you plan to 
embrace library management tools ala OSGi, but they are not a blocker 
for me.

I also think that adding documentation could take less time than all of 
this discussions and would be more user-friendly, but this is utopia 
existing only in my mind.

The most user friendly action we can do is releasing, so I'm happy with 
any solution you will agree upon as long as it will not delay the 
release (ok, we are not in hurry as we don't have a release manager, 
yet, so take your time).


PS: I repeat myself that discussion before commit and RTC does not worth 
too much here and really prefer CTR and vetos, because how we do stuff 
now is discuss a lot, find somehow an agreement, make a branch, have 
people review, commit and only then other people care of what you 
committed and complain. I'm not saying that this discussion is not 
good..but I hope people will remember this thing when I'll give up 
waiting weeks to form a consensus before committing code. If I committed 
this stuff as the first step we would be here discussing the same way, 
but at least I would have not lost time preparing a demonstrative 
branch, a proposal, a JIRA issue, collecting opinions and trying to find 
consensus. Is there a more agile way for us?

To unsubscribe, e-mail: server-dev-unsubscribe@james.apache.org
For additional commands, e-mail: server-dev-help@james.apache.org

View raw message