[development] Download stats for October

[Chris Kennedy]

Comparing cvs downloads of drupal and all contrib modules isn't
worthwhile for a number of reasons:

1. They're used for different purposes - drupal cvs is primarily for
testing where as cvs versions of modules tend to have the most bugfixes
and functionality. And as you state, drupal-cvs is often (usually?)
updated directly though CVS rather than a download. And this October
-cvs version downloads could simply reflect additional testing on Drupal
5.0 module ports.

2. They reflect different versions - drupal-cvs for October is known to
be Drupal 5.0, but for modules -cvs might be a 4.6, 4.7, or 5.0 version.

3. One is unitary (Drupal) whereas the other is aggregate (modules).
There are about 550 modules tagged for drupal-4-7 (thanks dopry), so I
don't see a 16-folder difference in cvs downloads as an inherent
disproportionality.

There are better ways to examine what you are getting at:

a. Compare cvs / 4.7 download ratios on a per-module basis and examine
the distribution for modules with both variables. This would be easy
with a bit of string manipulation on the download list.

b. Count the percentage of modules that set cvs/head as their default
download on d.o. (or do a db query).

c. Look at the distribution of time between the latest -cvs release and
DRUPAL-4-7 release on a per-module basis, though this would need to be
tweaked with the shift to Drupal 5.0.

Great stats summary otherwise - the reasoning behind this issue (but not
necessarily the conclusion) was the only thing I disagreed with.

-Chris

Greg Knaddison - GVS wrote:
> On 11/15/06, Gary Feldman <dpal_gaf_devel at marsdome.com> wrote:
>> Greg Knaddison - GVS wrote:
>> > ...
>> > +++Downloads by tag:
>> >
>> > 4.7    439049
>> > cvs    27294
>> > 4.6    11190
>> > 4.5    234
>> > 5.0    46
>> >
>> > To me this indicates that module/theme maintainers need to do a better
>> > job of branching their modules especially when combined with the
>> > following chunk:
>> I don't follow this.  Is the reasoning that 11190 downloads for 4.6 is
>> high, suggesting that people have to use 4.6 modules for 4.7
>> installations?  There were nearly 2000 downloads of the drupal-4.6 tar
>> file, so the ratios don't support this (439049 is roughly 20 times the
>> number of 4.7.3 downloads, while 11190 is only 5 times the number of
>> drupal-4.6 downloads).  I'm not sure this reasoning has any validity, so
>> I'd like to know more about your conclusion.
> 
> Sure - let me give my logic in longer form:
> 
> 27,290 downloadeds happened for files (modules, themes, core,
> translations) for the CVS  revision.  Yet when you look at the numbers
> just below this section that you snipped out there were only 1,316
> downloads of Drupal Core CVS.
> 
> So, given that there were only approximately 1,316 installations of
> Drupal CVS, why are there so many downloads of Modules from the "CVS"
> tarball?
> 
> 1. A module that had a 4.6 branch and has had work in HEAD for 4.7 but
> doesn't have a 4.7 branch.  That's the most likely scenario that I've
> encountered.  A perfect example is the leaf theme:
> http://drupal.org/node/20591/release  HEAD works with 4.7, but results
> in a download that counts in the CVS pile.  That theme should ideally
> be branched for 4.7.
> 
> 2. A user went to the project download page, saw the cvs tarball as
> the default release, and they grabbed it even though they have
> Drupal4.6 or 4.7 installed.  This is something I have encountered
> pretty often in the issue queue, but it seems less likely.  I can't
> find any examples of these at the moment, perhaps the recent project
> work changed that, or perhaps michelle fixed them all.  Either
> way...it's a possible scenario.
> 
> 3. Some other slight variation on these two.  For example, a variation
> on scenario 2 is that a user goes to the project page, sees the 4.7
> version as the default release, sees "view other releases, checks
> them, sees a CVS release, assumes that the CVS release will work with
> 4.7 and has updates not available in 4.7 so they download the CVS
> version.  I have encountered this scenario in the issue queue as well.
> 
> Now, if I understand your point it's that people download multiple
> modules/core which is a good point.
> 
> If we use 4.7 as the baseline there were 39855 core downloads, 329387
> module downloads for a ratio of 8.26 modules downloaded/core download.
> 
> If we look at CVS there were 1316 core downloads, 20478 module
> downloads for a ratio of 15.56 modules downloaded/core download.  This
> looks to me like a disproportionate number.
> 
> Finally, looking back at 4.6 there were 3629 core downloads, 5419
> module downloads for a ratio of 1.49 modules downloaded/core download.
> However, due to the relative lack of popularity of 4.6 in general and
> the fact that most 4.6 installations are in maintenance mode (note the
> popularity of the PHPTemplate engine for 4.6) I think we should ignore
> 4.6 and only use 4.7 as the baseline.
> 
> So, perhaps that is faulty logic, but there it is.  I'd be happy to
> hear other explanations for the disproportionate ratio between
> module/core downloads for CVS.
> 
> Regards,
> Greg
>