On 2/13/19 1:44 PM, Andres Freund wrote:
> Hi,
>
> On 2019-02-13 13:01:46 -0500, Jonathan S. Katz wrote:
>> So, in anticipation of push back, I also want to provide some stats. I
>> looked at overall traffic since Jul 1, 2018 to yesterday. Traffic to any
>> of the release notes in general was a very small chunk of it (< 0.5%).
>
> Given the number of pages, that doesn't seem that small. I personally
> have hit broken /devel/release-xy.html links several times since the
> hasty removal.
It's an incredibly small amount relative to other parts of the site. By
the above measure, it's statistically insignificant even with alpha of .01
To give it more context, it makes up 1% of all pages that are viewed in
the documentation. Looking from Jul 1, 2018, in the top 100 viewed doc
pages, #91 was the PG11 release, and nothing else. Basically, the major
release pushes the views of the release notes way up by a significant
portion.
Also, this did hurt my workflow as well when putting together the press
release. Given the release notes were not in devel, I ended up building
the docs from the REL_11_STABLE branch to put it together (rather than
parsing the SGML manually). It was inconvenient, but IMV the benefits
far outweigh the inconvenience.
>
>
>> From a user experience perspective, does it stink that we're potentially
>> creating a bunch of 404s? Yes. Is it going to upset some people? Yes,
>> and I'm sympathetic to that. However, looking at the overall numbers, I
>> think we'll be ok, even if we don't have the redirects per Magnus'
>> suggestion downthread (which I will also reply to).
>
> We're going to break current/relase-xy.html links after every single
> major release this way. This seems absurd.
We mentioned a way we can fix this -- having a map from /current/ (and
if need be /devel) to the proper version of the old release notes can be
done with some regular expressions, as Magnus said. It should not be too
difficult to do. I may even be able to get a POC done later today.
> I'm quite pissed at how haphazardly this was done.
We had discussion on this for months, and after some idling, we proposed
a solution and worked through it quickly. Sometimes you just need to rip
the band-aid off.
The nice thing about web is that you can test your assumptions and
iterate much more quickly than other ways to deliver software (e.g. a
database).
Thanks,
Jonathan
