[QGIS-Developer] using conda for development

[Ari Meyer]

Yes, Greg, it would be interesting to know if this resulted in many actual
incompatibilities or QGIS has managed to just be lucky.  Even if there are
no API changes, implementation changes can certainly result in
regressions.  I'm not sure what percentage of open-source projects still
follow a ">= version X" policy for dependencies.  In the Java world, I
haven't seen many specify Maven or Gradle dependencies like that (or even
range-bound versioning), although they can.  The general policy that I've
seen is to always specify precise dependency versions.

On Sat, Aug 1, 2020 at 8:09 PM Greg Troxel <gdt at lexort.com> wrote:

> Ari Meyer <ari.meyer at gmail.com> writes:
>
> > Matthias, WADR, your statement seems, at least on the surface, to
> indicate
> > an incredibly fragile process.  How can you possibly maintain stability
> and
> > reproducibility if the dependencies can be changed at will?
>
> I can understand how you would say this, but I find your statement odd
> from an open source development norms viewpoint.   It is entirely
> typical for a project to specify a list of dependencies with a miniumu
> version (because of API changes), with the notion that if built against
> that version or later, all will be ok.  That leads to the application
> being judged buggy, if it doesn't cope, and the dependency being judged
> buggy, if it doesn't meet the interface specification.   Certainly
> hard-core testing will be against particular versions.  The history of
> open source projects is that usually things are ok with this fuzzy >=
> binding.
>
> In my little corner, qgis makes releases, and they are built within
> pkgsrc, which means against whatever version is in pkgsrc that minute.
> Aside from upstream instability of some packages, this is mostly ok.
>
> I have only been using qgis the last month or two, but have been lurking
> and thinking about packaging for much longer.  I am curious how many
> documented cases of incompatibilitiies due to versions of something --
> when the qgis build says both are acceptable -- there have been, and if
> in hindsight the bug is in qgis or the dependency.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/qgis-developer/attachments/20200802/56980a2f/attachment-0001.html>