Re: [Jython-dev] Bundled/shaded JARs

[Jim B. <jim...@py...>]

The socket-reboot branch, bringing back the memories!

Adam is absolutely right about the use of shading. Although in our defense,
when we started using jarjarbinks, it was the best available choice out
there to manage these dependencies. The world has moved on, and the recent
Gradle work will hopefully be a good bridge to that.

re these shaded module imports: I think the only thing here that might have
been introduced in this branch is working around the shading of included
libraries (netty, etc) from Python library code (socket, ssl, select).
There is certainly no attempt to be systematic about fixing other
dependencies like JNR, which can still see dependency hell if JRuby is also
installed.


On Mon, Feb 3, 2020 at 2:04 AM Jeff Allen <ja...@fa...> wrote:

> On 03/02/2020 08:45, Adam Burke wrote:
>
> In other codebases I have seen this done reactively as a “poor man’s
> dependency conflict resolution”. ie introduce renaming only where there are
> conflicts in transitive dependencies, allowing the use of two different
> versions of classes with the same full name.
>
> I don’t know the specific history in Jython, though.
>
> Here from the beginning:
> https://hg.python.org/jython/rev/107fe4a4c96b#l28.7, hence pinging Jim.
> Unfortunately, it didn't occur to me it might be the wrong thing to leave
> it so until after I cut a beta 3 (unpublished so far).
>
> Now wondering if a beta 4 might be necessary as it seems the sort of
> change that would need it, in case unshading produces the sort of conflict
> you're talking about.  :(
>
> Jeff
>
>
> Cheers
> Adam
>
> 在 2020年2月3日，下午6:36，Jeff Allen <ja...@fa...> <ja...@fa...>
> 写道：
>
> 
>
> When we install Jython we have bundled the contents of the JARs we depend
> on into ~/jython.jar . Some we name-translate:
>
> org/python/apache/commons/compress
> org/python/bouncycastle
>
> And some we just copy:
>
> com/ziclix/python/sql
> jnr/ffi
>
> What drives one choice or the other?
>
> Testing after installation, I get a test failure from test_ssl (Unable to
> create OpenSSL PBDKF: PBKDF-OpenSSL SecretKeyFactory not available), where
> I suspect the shading is the cause. Quite likely, something is loaded by
> its unshaded class name, since if I place unshaded BC JARs on the path the
> test passes.
>
> We seem to have gone to some lengths in _sslcerts.py, for example, to use
> the shaded classes, but must we shade it?
>
> Jeff
>
> --
> Jeff Allen
>
> _______________________________________________
> Jython-dev mailing list
> Jyt...@li...
> https://lists.sourceforge.net/lists/listinfo/jython-dev
>
>