[prev in list] [next in list] [prev in thread] [next in thread] 

List:       pkgsrc-users
Subject:    Re: Python package issues
From:       Mike Pumford <mpumford () mudcovered ! org ! uk>
Date:       2019-05-25 21:26:51
Message-ID: 5c334f62-02a5-6749-57c8-a41aa52ba43a () mudcovered ! org ! uk
[Download RAW message or body]



On 25/05/2019 21:54, Greg Troxel wrote:
> Mike Pumford <mpumford@mudcovered.org.uk> writes:
> 
>> Sometimes you can't. If you have a package chain that depends on the
>> py27 and py37 version of the same package you can't do a make clean in
>> between. Only fix here is for pkgsrc to do a make clean after make
>> install for packages being installed as dependencies.
> 
> The other change, that has has some discussion, is to make WRKDIR be
> work-$PKGNAME or something like that, for all packages that have
> versions.
>
That's just as valid a solution.

> It is an interesting suggestion to have the dependency call be "make
> package-install clean" instead of just "make package-install".  We have
> generally avoided implicit cleans, but I'm not sure that's helpful on
> balance.
> 
I can see why it might be useful during package development. In that 
scenario you might be trying to figure out why a dependency hasn't 
provided a file it was supposed to. If the work directory for it is gone 
that will become more difficult.

Trying to debug issues with packages built using pkg_comp can be 
frustrating becuase of its habit of forcibly cleaning up at the end of 
the build of each package set so the no-clean policy does have some merit.

Having an option to turn it on might be useful for users of pkgsrc 
rather than developers though.

>> I ran into a similar issue trying to build xfce4 in a chroot. gtk2
>> blew up as its dependency chain included python 2.7 and python 3.7
>> versions of the same package.
>>
>> I've still got the error log from the build in my jenkins server if
>> you want complete output.
> 
> No need; the problems are understood in theory :-)
>That's fine. My xfce4 build was experimental and I'd not really had that 
much time to pursue the fallout.
> loop:
>    make package-install
>    # observe failure
>    rm -rf /usr/pkgsrc/*/*/work
>    goto loop
> 
That's somewhat more difficult to do when you aren't building packages 
manually but are instead relying on something like pkg_comp. I'm 
guessing pbulk does things differently enough not to fall down this 
hole. I might have a quick stab at making the dependency package-install 
step add a clean as well. If I could find a simple way to combine 
libkver with pbulk I'd have switched away from pkg_comp by now.

Mike
[prev in list] [next in list] [prev in thread] [next in thread]