[aur-general] TU Meeting

Allan McRae allan at archlinux.org
Sat Nov 29 04:36:19 EST 2008

Well, given you don't wait the few hours to put this forward at the TU 
meeting, I won't wait to respond to you.

w9ya wrote:
> Hi all;
> Some quick comments/corrections to Allan's Wiki page;
> - His first question/answer about the purpose of the Community Repo; 
> <- The AUR was not in existence in ANY form for several years AFTER 
> the introduction of TU **BINARY** repos. While Allan (and others) may 
> truly believe that the community repo is designed strictly to be a 
> place for popular AUR contributions there is NO historical basis for 
> this belief. i.e. We the TU's, as WHOLE group, have *never* determined 
> this to be a sole basis OR requirement for binary packaging. Nor, in 
> the past when this has come up (and it has a few times), have we EVER 
> decided to make such a decision, as it was considered to be a bad idea 
> to do so for a variety of still pertinent reasons.

OK.  So lets look at the revision of the AUR Truted User Guidelines from 
July 2005:

In the description of a TU: "He maintains popular packages".  That is 
the oldest (and unchanged since then) definition of a TU we have to go by. 

For those of us that are historically unaware, can you outline your 
"still pertinent reasons".  In all your previous emails saying similar 
things I saw no reason other that "it was never the way".  Sure it 
wasn't the way in the start (as there was no AUR) but things change.

> AS SUCH, Allan's premise is flawed as is everything else on the page 
> is based on this erroneous premise of purpose.
> (BTW, I have been told such historical examinations are "regressive" 
> and as such NOT "progressive". I will encourage you all to decide if 
> such an understanding is "regressive" or really just a useful way to 
> gain knowledge and understanding and prevent mistakes.)
> - Allan goes on to discuss what a "popular" package is. There is 
> absolutely no way to determine what 3, or 20, or what any number of 
> votes truly represents for a variety or reasons. I have discussed just 
> why there can be no correlation in a previous email. Prior to sending 
> that email I asked for such a correlation and received none. I then 
> wrote that email where I outline the flawed mathematics, I have yet to 
> see a rebuttal. There is NO way to tabulate and correlate the votes as 
> to representing even a percentage of users. Do NOT let Allan or anyone 
> else tell you otherwise. It simply is NOT true at this time.

I gave a link to a plot on the wiki page showing the correlation between 
number of votes and usage from pkgstats for my package.  The correlation 
is visibly high, with the exception of packages included as 
dependencies.  As we have no better numbers available to us, and the two 
numbers we do have agree reasonable well, we might as well use what we have.

> TO WIT; whether it is 3 votes (the original minimum suggested) or 20 
> votes; they are merely a number with NO MEANING. SO why should we 
> place ANY **ARTIFICIAL** meaning to them ? AND why are we using them 
> at all ?
> - It is also worthy to note that in the course of a week or two since 
> this minimum number was first proposed it has changed from a 
> requirement of 3 votes to 20 votes. What will the number be next month 
> ? Six months from now ?

The minimum of three votes is clearly flawed if you look at the pkgstats 
usage and compare the number of votes.  In fact, I linked to another 
plot which clearly shows this.  The number 20 was a guideline that was 
around when I first joined as a TU and if you look at the pkgstats 
results and try separating packages with more or less that 1% usage, the 
optimal number of votes arrives very near to 20 so I stuck with it.

> Why should any TU even want to contribute new and interesting BINARIES 
> when the number can change so quickly and by a small number of 
> "leaders" deciding for him/her ? Why should ANY TU spend time 
> *properly* vetting an AUR contribution through adoption only to have 
> it removed at a future date because it no longer had the require 
> number of votes ?

As I said in the wiki page, I was not proposing the enforcement of any 
removal based on these numbers.  I only proposed guidelines for getting 
a package into the [community] repo.

> FURTHER; when have we EVER let a small number of TUs make any 
> decisions for the rest of the TUs ? IS it wise to change the 
> successful TU system into merely a training ground for future DEVs and 
> all of the TUs merely "junior" DEVs grinding out packages because they 
> receive votes ? Where is the creativity in that ?

We have never let a small number of TUs decide.  In fact, any change are 
always required to reach quorum and a percentage of the vote as given in 
the bylaws.

> - Finally, but far from the least important consideration, is that we 
> are being asked to do this because of a lack of server resources. I 
> would like to remind everyone that this proposal will only delay the 
> day until this minimum number of votes is increased OR the servers 
> will need to be upgraded. I for one would rather like to contribute 
> some real hard cash and simply do the necessary infrastructure 
> upgrades and NOT set such a precedence of paradigm changes in a very 
> successful system. Changes that will be hard or virtually impossible 
> to remove once they are adopted, more especially because they will NOT 
> solve a resources problem for any useful length of time. Can ANYONE 
> truthfully assert that these changes will alleviate ANY resource 
> problems for any real length of time ? The only correct answer is no. 
> And so the folks presenting the TUs with this decision have said as 
> much in prior emails.

Resources are limited, no matter how much money you throw at a problem.  
Every extra package uses bandwidth and disk space for both us and our 
mirrors.  And if you are only building for i686, it costs x86_64 TUs 
time to build these packages. 

We are talking about optimizing the use of our resources.  Out of the 
~2400 unique IP who submitted their package statistics, 97 packages in 
[community] are used by no-one.  That cannot be optimal when plenty of 
packages in unsupported are having usages between 1 and 5%.

> ****** Oh yeah, DO NOT FORGET that there are no alternative proposals 
> to remedy the server problems because they have NOT been outlined in 
> enough detail to determine if this is the best or even a good proposal 
> to alleviate these vaguely declared problems. ******** i.e. WHAT 
> problems, or WHAT nature, and WHY are they NOW a problem ?

It is an optimal use of limited resources issue.  Find a way to convince 
me (and others) that serving packages out that from all indications 
no-one or very few people use is optimal.

> Thanks for reading this far. I hope every TU really asks themselves 
> whether this has been truly well thought out or not. Ask the tough 
> questions. Seek an understanding of just how this will help and 
> whether it is truly of merit or merely the beginning of other changes 
> that will result in a paradigm shift in our VERY successful and truly 
> unique TU/AUR system.

Which is why we are having a meeting.  So we can discuss the relative 
merits/demerits of the proposal.

> Very best regards;
> Bob Finch

Who still top posts....

> On Fri, Nov 28, 2008 at 8:15 PM, Allan McRae <allan at archlinux.org 
> <mailto:allan at archlinux.org>> wrote:
>     The first TU meeting in a long time (over a year!) is on at 9.00pm
>     GMT, Sunday 30 November.  Afternoon US time, early Monday morning
>     for me...  we will keep a log for those who can't make it then.
>      Anybody needing the TU channel key, send me an email.
>     I have made a page[1] with what will be the main discussion topic
>     and what I propose to improve the community repo.  Feel free to
>     make additions to the page (and sign them so we can tell who wrote
>     what), especially the final proposal.
>     Talk to you all then.
>     Allan
>     [1]
>     http://wiki.archlinux.org/index.php/User:Allan/Community_Repo_and_Pkgstats

More information about the aur-general mailing list