Re: Windows build on the buildbot

[Friedrich Beckmann]

Good morning Harry!

> The missing of adwaita en sourceview (btw why not 3.24.11?) is a known issue. 
>  What is missing can change with every openSUSE release.  :-(

I used the same version that is in opensuse 15.2. Mindaugas maintains pspp on 
opensuse. Maybe somebody wants to maintain a opensuse mingw repo with the 
missing parts like Mindaugas for pspp:

https://build.opensuse.org/package/show/home:embar-/PSPP-master

such that all required packages end up in:

https://build.opensuse.org/project/show/windows:mingw

> I didn't see where you install libiconv and spreadsheet-widget. I would 
> recommend to do everything by scripts. It is somehow a documentation and much 
> better if you have to do it again for lets say openSUSE 16.0

The build control on the buildbot is split. 

1. Toplevel is the master.cfg from buildbot: 
https://github.com/fredowski/pspp-buildbot/blob/master/master/master.cfg#L136
2. That calls run_win.sh: 
https://github.com/fredowski/pspp-buildbot/blob/master/run_win.sh
    => Build spreadsheet widget via win/buildssw-win.sh
    => Run win/buildpspp4windows.pl

I take libiconv from opensuse and that is already in the prepared build 
environment, i.e. it is not installed in every
build run. The build environment is created with 

create_win_vm.sh
  => prepare_win.sh (running inside the lxc container)

So libiconv is here: 
https://github.com/fredowski/pspp-buildbot/blob/master/prepare_win.sh#L38 and
spread-sheet is here: 
https://github.com/fredowski/pspp-buildbot/blob/master/win/buildssw-win.sh

I also think that without a scripted setup you are lost.

> And good to chmod not to 777. :-)

I think your script does the sudo chmod. The pspp user inside the lxc container
does sudo without password...

> You didn't handle Postgres I think. But I doubt it worked in my setup. I 
> never tested it and nobody ever complained about it.

Yes, I don’t care about that. I wonder if people even use the pspp terminal 
version…

> I fiddled a bit to build the 1.4.0 release - in the end I renamed
> the files a bit and that is what you find in downloads.
> 
> Yes the release version is always a bit more work.  And actually nobody  
> whats them because they need the translations too which is included in later 
> test versions.  I usually take a version af a few weeks later as teh released 
> version.

Yes, there are problems with the translations in the release. It is a pity 
because we normally try to test the release especially the gui a bit more. This 
time John waited with some major code changes just after the release.

> It is hard to test. And you never know what people around are doing.  I 
> usually do a smoke test and run a syntax script. If that seems to be ok I 
> make them available for download. I have no automated regression tests.
>  
> You have a great script! That must have been hard work to figure out
> all the settings and the flow for the complete installer with this nsis
> and the gdb debugger. I found very little about doing such a windows build.
> 
> Well it is a work of about 10 years. Every problem solved and put in the 
> scripts. 

Really great!

> The build is done by copying the template opensuse virtual machine (lxc 
> container)
> which is disposed after the build. As you said the environment is frozen but
> it should be possible to update the packages in the build machine during the
> build. 
> 
> Be careful even with that. I have seen it often that the dll hell traps in 
> when you update something.  If possible create a complete new environment if 
> an update is needed. A new environment is called a new generation by me. 
> Users have to remove their older installation before installing. When in the 
> same generation, installing on top is ok.

I have the same experience for the MacOS application build. It became better 
since macports does more integration tests.

> It seems the build has not correctly worked today. 

Could be - I tried to do something about the Wformat warnings due to %zu format 
and added -D__USE_MINGW_ANSI_STDIO=1 but at least the warnings remain there.

> As far as I can see the buildpspp4windows.pl has only very minor changes.  
> Most important you changed webdir and not uploading the result as far as I 
> can see.

Yes, your script worked right away!

> Is it your intention to run this continuously?  It looks as if it is not to 
> be much work for you. 

Well, my plan is only to have the windows build on the buildbot to see if 
anything breaks the build there.

http://caeis.etech.fh-augsburg.de:8010/#/builders/7

The buildbot provides a nightly build of your windows installer:

https://caeis.etech.fh-augsburg.de/downloads/

But I only have windows 7 32 bit in a vmware vm. So I am not testing this. The 
missing regression results in pure build test. But to figure out „funny 
characters“ somebody must run this on Windows.

Fritz