As many of you know, Maven makes a lot of assumptions and does a fair amount of
work in each assumption. So, for example, when you run "mvn install", it runs
several steps including the build and copies it to the local repo. Imagine the
number of lines of Ant code required to do the same.
I am told that the app that you run when you go to the ASF Maven Staging Repo
in your browser (repository.a.o) also has a pile of assumptions. So when we
copy artifacts up there and tell the app to "close" a staging repository, it
runs several steps that would amount to a huge pile of Ant scripting. And
later, when it is time to push it all to Maven Central, it can do that as well.
Similarly, Maven's "release" plugin makes assumptions that the ASF Maven
Staging Repo expects. So it packages up sources, makes checksums, signs
artifacts and pushes them to the staging repo. That would also require a lot
of Ant scripting to do, although some of it, like a source artifact per-jar,
IMO, is not absolutely required.
So, that's why I think it is the right thing, assuming we want to publish Maven
artifacts for our releases, which I think we do want to do. It is far less
work on the RM to maintain less verbose Ant scripts which are already pretty
long. The assumption that you have reliable high-speed internet connections
shouldn't be that hard to meet these days, but the Repo server did seem to
care, or maybe it was my home router that got overwhelmed and dropped packets.
Not sure. All I know is that I saw it work without tweaking any network
settings.
HTH,
-Alex
On 11/15/18, 11:37 AM, "OmPrakash Muppirala" <bigosma...@gmail.com> wrote:
On Thu, Nov 15, 2018 at 2:02 AM Alex Harui <aha...@adobe.com.invalid> wrote:
> To be clear: the process DOES NOT upload artifacts to Maven Central. It
> uploads to the ASF staging repo. Only after we approve the release do the
> artifacts get pushed to Maven Central. One of the RMs got an error
> DOWNLOADING a Maven plugin from Maven Central. There is nothing wrong
with
> doing that.
>
Sorry, I misspoke. But the question remains. Why do we need to upload it
to the staging server during building or testing a release candidate? Can
we not just use the local m2 repository for this?
In other words, what aspect of the release are we actually testing by
uploading stuff to the staging repo?
>
> Any volunteer is welcome to re-do the release scripts. What we have now
> is what I think was the simplest sequence for creating both Maven and
> non-Maven artifacts. Maven was the number 1 asked for feature in Adobe
> Flex. IIRC, it had 100 votes and 2nd place was 12 votes. So we support
> Maven in hopes that the customers we want to attract are using Maven, plus
> at least Carlos prefers to use Maven.
>
> Just like any other code or script in Royale, there can be bugs and room
> for improvement in the release process. I found and fixed some issues,
and
> hopefully the next RM will try the scripts after a month or so instead of
6
> months so fewer things will have become stale and broken. In fact, I
> think I will say that RC2 is my last RC for a while. If we need an RC3,
> someone else should create it since the know the scripts recently worked
> for me. And since it is fresh in my mind, I might be able to help the
next
> RM better. I think the problems Piotr and Yishay hit have been addressed.
> The only remaining issue is upload/download issues.
>
Thanks for getting the scripts to this shape!
Regards,
Om
>
> Thanks,
> -Alex
>
> On 11/14/18, 10:36 PM, "OmPrakash Muppirala" <bigosma...@gmail.com>
> wrote:
>
> Personally, I am not sure why we are uploading stuff to maven central
> as
> part of the RC process. Can we simply bundle the source files and
> create a
> release candidate? Folks who want to test maven can run the maven
> scripts
> locally which would copy stuff to the local .m2 repo instead.
> That way we dont force maven on folks who dont need it.
>
> Thoughts?
>
> Thanks,
> Om
>
>
>
> On Wed, Nov 14, 2018 at 12:54 PM Carlos Rovira <
> carlosrov...@apache.org>
> wrote:
>
> > Hi Alex,
> >
> > first congrats to reach to this point. After many months is great to
> > finally can release :)
> >
> > I forked the thread to avoid mixing topics and let the discuss
> thread to
> > just discuss the RC2
> >
> > I'll try RC2 soon, but first I want to talk about the inherent
> problem. I
> > know we are all here making a huge effort to make Royale work and
> shine,
> > and be release manager is to spend many hours working in that task,
> so
> > first I want to express as well a huge recognition to all RMs that
> tried to
> > make this happen before you. I still didn't try to make myself a
> release,
> > so I'm talking without the knowledge you all guys have, so sorry if
> I say
> > something inconvenient now, but just want to talk about the elephant
> in the
> > room, so we can improve the process from now own.
> >
> > My perception, from a person that is still outside the release
> process,
> > still does not have all the knowledge and never was a RM is that the
> > process is not reliable. Again hope nobody feels that talk about
> this is
> > any kind of attack, just that if we agree on that, we could try to
> improve
> > for the better. Right now seems that only Alex can make a release
> happen
> > due to his machine configuration. I assume all RMs execute the RM
> > instructions, so the problem is not related to the instructions, but
> seems
> > more to the nature of how we build.
> >
> > So, my question is, what we could change to ensure scripts and
> process are
> > reliable in all machines and doable by any of us?
> > Is the process a mix of maven and ant what is making this so hard to
> do?
> > Just a comment, could be the nature of ant (more single separated
> scripting
> > based on computer environment config vs structured building based on
> > building framework and servers) a problem?
> >
> > I want to be a RM soon (hope to be available for this task around
> Feb-Mar
> > 2019), so its clear I still can say a word about the process just
> start
> > this thread and hoping no one feels this as an attack (since I
> recognize
> > the huge effort done to reach this point), but just a way to start
> trying
> > to think with open mind and see that we have at least a thing to
> solve so
> > any of us can make the same as Alex do
> >
> > Thanks in advance for your thoughts :)
> >
> > Carlos
> >
> >
> >
> >
> >
> > El mié., 14 nov. 2018 a las 19:46, Alex Harui (<aha...@earthlink.net
> >)
> > escribió:
> >
> > > This is the discussion thread. RC1 was never put up for vote
> because
> > > there
> > > were issues with the notice files and build scripts when running
> from the
> > > packages and not the repo.
> > >
> > > Interestingly, the scripts ran just fine for me on my home Windows
> > > computer.
> > > I did not have the problems Om, Piotr, and Yishay had when
> uploading or
> > > downloading Maven artifacts. Except when I tried watching videos
> while
> > > waiting for the scripts to run (the "maven" target takes 2 hours
in
> > > Powershell, the "main" target took over an hour). Then I would
> see
> > > uploads
> > > "hang". So I would recommend to any future RMs to try to make
> sure the
> > > network is not otherwise too busy when running these scripts.
> > >
> > > Changes in this RC include:
> > >
> > >
> > >
> > > Thanks,
> > > Alex Harui
> > >
> > >
> >
> > --
> > Carlos Rovira
> >
>
https://na01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fabout.me%2Fcarlosrovira&data=02%7C01%7Caharui%40adobe.com%7Cab398626cd724476962608d64b31c383%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C636779074468579898&sdata=a9%2F3296HxhUsFeGuPyZPaH%2BJOXrU2PxvMYgi%2FMaxsZs%3D&reserved=0
> >
>
>
>
