gmane.comp.file-systems.zero-install.devel

Version 1.7.1 for Windows released

Bastian Eicher — 2012-05-23T16:35:43

Version 1.7.1 of the Windows version of Zero Install has been released. Changes in this version include:
 * Added support for case-insensitive Windows file paths as arguments for "0install add/remove-feed" and "0install add/integrate/remove-app"
 * Fixed crash when main window is closed before an integration dialog
 * Fixed crash when file system access is denied when deleting an implementation
 * Fixed sporadic crash after successfully launching an implementation
Existing installations will update automatically after confirming a dialog box.
You can also trigger an immediate update using the command line '0install self-update'.
For new installations please download here: http://0install.de/downloads/

Release notes: http://0install.de/news/v1-7-1-released/


------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions

Package dependency and multilib

Michel Alexandre Salim — 2012-05-23T09:40:29

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I'm using 0install on a Fedora x86_64 system, and I notice that when
installing an 'i386' feed with package dependencies (in this case,
http://hircus.multics.org/interfaces/2011/skype.xml), 0install cannot
resolve the package dependencies if they're not already installed.

To try:

$ yum remove glibc.i686 # make sure no i686 packages are installed
$ 0alias skype http://hircus.multics.org/interfaces/2011/skype.xml

This will fail, with the Version field of the dependencies
(alsa-plugins-pulseaudio, libXss, libXv, qt-x11) being marked in red
as "(problem)" - No suitable version was found. Double-click here to
find out why.

Double-clicking shows, for example for libXss, the following:

Version | Released | Stability | Fetch     | Arch     | Lang | Notes
1.2.2-1 | -        | packaged  | (package) | *-x86_64 | -    | None

I'm not entirely sure if the following "just works" in older releases,
in which case this is a regression, or if I've previously always had
those

suggestion: 0alias --manpage should justfallback to calling man

Tim Cuthbertson — 2012-05-22T02:24:17

with `man` aliased to "0alias --manpage --", and a vim alias (which is
a custom vim feed that relies on a package implementation)

$ man vim
No matching manpage was found for 'vim'
(http://gfxmonk.net/dist/0install/vim-custom.xml)

$ \man vim
(a manpage appears)

You might run the risk of manning the wrong thing (if the 0install
feed doesn't correspond to the same named thing on your system), but
that's unlikely. To combat that we could add stuff to expose man-pages
in <package-implementation>s, but I'm not sure it's worth the effort.

Cheers,
 - Tim.

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Optional sub-packages

Dave Abrahams — 2012-05-21T17:09:10

Over at Ryppl we've been starting to wish more and more for optional
sub-packages.  The key area in which this could cause incompatibility
with ZI is that we think we need the optional parts to install into the
same directory tree as the other other parts, with some identifiable
relationships.  That would seem to wreak havoc with the cache,
manifests, and digests, at least as they currently work.  What's ZI's
thinking on this?

Thanks,

zeroinstall on homebrew

Tim Cuthbertson — 2012-05-21T11:47:02

Just a heads up: a while back [1] Anders was trying to get zeroinstall
into homebrew, but they were convinced that people should use a
combination of `brew` and `pip` to install pygtk (despite the fact
this didn't actually work). I noticed they seem to have changed their
minds and now there's a pygtk formula [2]. So it should now be
possible to submit a zeroinstall formula if you (or anyone else) is
still keen on this.

[1] http://old.nabble.com/Zero-Install-%28Injector%29-for-Homebrew-%28Mac-OS-X%29-td30713609.html
[2] https://github.com/mxcl/homebrew/blob/master/Library/Formula/pygtk.rb

Cheers,
 - Tim.

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/

Windows portability series

Dave Abrahams — 2012-05-19T15:14:54

This is the third time I've submitted most of these changes.  Is there
something wrong with the way I'm going about it?

[PATCH 01/14] Windows compatibility for setup.py
[PATCH 02/14] .gitignore build detritus
[PATCH 03/14] Adjust tests for moved top-level scripts
[PATCH 04/14] Low-hanging fruit: replace os.spawnlp(os.P_WAIT, ... )
[PATCH 05/14] Make sure invocations of 0launch from testrun.py still
[PATCH 06/14] Always generate manifests using Unix path separators
[PATCH 07/14] Fixed usage of os.rename that only worked in Posix
[PATCH 08/14] Respect XDG_* environment variables even on Windows
[PATCH 09/14] Make sure NamedTemporaryFiles are closed before trying
[PATCH 10/14] Make sure target is gone before renaming files on
[PATCH 11/14] Use Popen instead of fork() for the testing server
[PATCH 12/14] Replace a fork/exec pair in unpack.py with
[PATCH 13/14] Open files in binary mode for Windows portability
[PATCH 14/14] Get ZI into the PYTHONPATH of the local feed

-------------------------------------------

0install 1.8

Thomas Leonard — 2012-05-19T10:41:17

0install 1.8 has been released:

 http://0install.net/injector.html


New features

- Added "Explain this decision" menu item to the Versions dialog box.
You can use this to find out why a particular version wasn't selected
(e.g. because it conflicts with something else, is marked as testing,
etc).

- Added new "POSIX" architecture that matches everything except Windows.

- SSL support: Use the "certifi" library's certificate bundles if
available (Dave Abrahams).

- Limit number of concurrent downloads from a single site. Opening a
large number of simultaneous connections causes problems for some
networks and servers.

- Added two-argument form of add-feed.


Bug fixes / improvements

- Use terminal instead of xterm, when running with quartz (Anders F Bjorklund).

- Try x-terminal-emulator first for <needs-terminal>. Reported by Dave Abrahams.

- Fixed mirror code to work on Windows. Was using os.path.join to
create URLs. Reported by Dave Abrahams.

- Fixed handling of missing local feeds (was broken in 0ins

More connection pooling?

Dave Abrahams — 2012-05-16T04:54:39

When I download feeds with refresh for
http://ryppl.github.com/feeds/boost/bind-src.xml it looks like we are
seeing another logjam of connections to the server while it tries to
fetch all the dependency feeds.  I'm not sure that's the culprit, but it
sure seems ridiculously slow.  Thomas, do you think you could add
connection pooling for that case also?

Thanks,

A little guidance

Dave Abrahams — 2012-05-15T20:19:00

I want to make 0compile on windows do all of its archive downloads by
invoking the windows-ported 0install tool, i.e. "0install download
<whatever>".  I think this is quite doable and I started crawling
through the source looking for where to hook the code... but I figured
someone more familiar with the implementation could save me lots of
time.

I was looking at two strategies:

1. Override some methods of fetch.Fetcher.  If that's the best approach
   I'd like to know which ones to override.

2. Add a RetrievalMethod that's always considered "best" and delegates
   the work to the windows 0install command.  If that's the best
   approach I'd like to know what's supposed to be in a RetrievalMethod
   (there's really nothing in the API docs) and any details of the
   0install command it should invoke.

If there's a better strategy of course, I'm all ears.

Thanks in advance!

An apparent solver/driver limitation I'd liketo overcome

Dave Abrahams — 2012-05-15T12:20:54

When the solver finds a dependency for which there's only a source
feed, it fails.  I'd like it to assume the source feed can be built,
and thus act as though the necessary binary is available.  Is there a
way to do that?

Dealing with hardcoded paths, intercepting system calls in general: sysintercept

Tim Diels — 2012-05-15T10:33:25

I have a tool under way that will help deal with hardcoded paths on 
windows. This should make it easier for (haskell, ...) programs to be 
used with zeroinstall.

The tool: sysintercept: https://github.com/limyreth/sysintercept

I succesfully managed to redirect a file read by a haskell program of 
a.txt to b.txt. I should be able to extend this to replace the cabal 
prefix so these binaries can be relocated. In fact, I'll extend it to be 
able to replace any path by regex replacement.

To distribute a (haskell) binary with it, you'd use sysintercept as a 
runner. Sysintercept launches the program, injects a dll into the child 
process, does some patching of kernel32 and then intercepts all system 
calls related to file paths. The overhead of all this is mostly a couple 
of ifs per system call, there's no slow emulation.

The tool is currently still unusable as it's just a hardcoded test for 
this one haskell program. I'm busy with school at the moment, so it 
probably won't be ready any time soon, just tho

command= attribute for

Dave Abrahams — 2012-05-14T06:41:03

I'd like to be able to express that the 'foo' command of package A
depends on package B and all of B's 'foo' command requirements.  AFAICT
that isn't possible now.  Could we add a command= attribute to enable
saying such a thing?

Experimental "app" support

Thomas Leonard — 2012-05-13T13:51:12

I've created an "apps" branch in the 0install Git repository with
experimental app support.

Apps work a bit like aliases, e.g.

$ 0install add rox http://rox.sourceforge.net/2005/interfaces/ROX-Filer
$ rox --version
ROX-Filer 2.11

The main difference is that apps store their current selections (in
~/.config/0install.net/apps/rox/selections.xml in this case). This
means that they start faster, because the solver isn't needed:

$ 0alias rox-alias http://rox.sourceforge.net/2005/interfaces/ROX-Filer
$ 0install add rox-app http://rox.sourceforge.net/2005/interfaces/ROX-Filer

$ time rox-alias --version > /dev/null
rox-alias --version > /dev/null  0.12s user 0.02s system 91% cpu 0.144 total

$ time rox-app --version > /dev/null
rox-app --version > /dev/null  0.06s user 0.02s system 92% cpu 0.082 total

When run, they still trigger a background update if they haven't been
updated for a while, and you can also update them manually:

$ 0install update rox
No updates found. Continuing with version 2.11.

They also

ZI path case sensitivity

Dave Abrahams — 2012-05-12T23:54:27

I think someone else reported a problem like this recently:

--8<---------------cut here---------------start------------->8---
Z:\src\ryppl\zi-win>0install remove-feed z:\Public\vmshare\zi-win\ZeroInstall.xml
The feed was not registered for any interfaces.

Z:\src\ryppl\zi-win>0install remove-feed Z:\Public\vmshare\zi-win\ZeroInstall.xml
The feed was successfully unregistered for the following interfaces:
http://0install.net/2007/interfaces/ZeroInstall.xml
--8<---------------cut here---------------end--------------->8---

It seems like 0install should be doing case-insensitive comparison here?

Windows 0compile downloading outdatedzeroinstall.

Dave Abrahams — 2012-05-12T13:16:59

This is  with the latest 0install and 0compile on windows.  I can't
quite understand why it would be downloading zeroinstall-1.6.  Can
anyone explain?

Thanks,

--8<---------------cut here---------------start------------->8---
C:\Users\dave\AppData>0install --version
Zero Install Command CLI 1.7.0.0
Copyright 2010-2012 Bastian Eicher et al.
This program comes with ABSOLUTELY NO WARRANTY, to the extent permitted by law.
You may redistribute copies of this program under the terms of the GNU Lesser General Public License.

C:\Users\dave\AppData>0compile autocompile http://0install.net/tests/Hello-scons.xml
Downloading http://downloads.sourceforge.net/project/zero-install/0compile/0.28/0compile-0.28.tar.bz2...
[====================] Complete
Extracting archive (0install-fetcher-v1mbfqyd.0tm)...
[====================] Complete
Generating sha256 manifest (r0klzcyf.zog)...
[====================] Complete
Downloading http://downloads.sourceforge.net/project/zero-install/0publish/0.20/0publish-0.20.tar.bz2...
[=====

0compile 0.29

Thomas Leonard — 2012-05-12T11:49:39

0compile 0.29 has been released:

 http://0install.net/0compile.html

This is a bug-fix release:

- Fixed autocompile on 64-bit machines. Was broken by 50a517b95717,
which added an extra value to the end of uname. This caused
autocompile to think that the host was always 32-bit, and therefore
incompatible with the 64-bit binaries that it itself created. Reported
by Dave Abrahams.

- Fixed registering of local feeds with autocompile. Register the new
autocompiled implementation against the interface we wanted to build,
which is not necessarily the one in the <feed-for>.

- More verbose display of autocompile plans.

- Cleanup: use Driver instead of Policy.

- Cleanup: pass Selections rather than a Policy.

- Code cleanup: use self.config instead of policy.config.

Use of element

Dave Abrahams — 2012-05-12T04:57:17

What's the expectation w.r.t. feed-wide metadata like <description> when
<feed> is used to delegate implementations to another feed?  Should all
the feeds match?  Should all but the "primary" feed omit the
information?  Should "secondary" feeds use <feed-for>?  The feed format
doc seems to imply they should, but so far I know <feed-for> as being
useful for local feeds... which use is not described in that document.

Thanks,

Windows version reading valid feed as garbage

Dave Abrahams — 2012-05-11T22:41:34

Please see
https://gist.github.com/babede36359c9ed3f012#L115

Also, I note a very suspicious-looking mixture of forward- and
back-slashes beginning here:
https://gist.github.com/babede36359c9ed3f012#L133

I've been unable to track down where either of these things comes from.
Help?

Thanks,

Why does ZI think I'm in off-line mode?

Dave Abrahams — 2012-05-11T21:14:17

I can't understand this.  Can someone help explain please?

0install says I'm online...

--8<---------------cut here---------------start------------->8---
$ DISPLAY= 0install config -v    
INFO:root:Running 0install 1.7 ['-v']; Python 2.7.3 (default, Apr 15 2012, 01:58:13) 
[GCC 4.2.1 Compatible Apple Clang 3.0 (tags/Apple/clang-211.10.1)]
help_with_testing = True
auto_approve_keys = True
freshness = 1d
network_use = full
--8<---------------cut here---------------end--------------->8---


... but then complains I'm off-line!

--8<---------------cut here---------------start------------->8---
$ DISPLAY= 0install -v update -r http://ryppl.github.com/feeds/ryppl/ryppl.xml 
INFO:root:Running 0install 1.7 ['-v', '-r', 'http://ryppl.github.com/feeds/ryppl/ryppl.xml']; Python 2.7.3 (default, Apr 15 2012, 01:58:13) 
[GCC 4.2.1 Compatible Apple Clang 3.0 (tags/Apple/clang-211.10.1)]
INFO:root:No run <command> in http://ryppl.github.com/feeds/ryppl/ryppl.xml
INFO:root:No run <command> in http://ryppl.github.com/feeds/

Windows version: Failed to check GPG signature

Dave Abrahams — 2012-05-11T14:40:00

I'm at a loss to explain this problem.  Any help?  It works on my Mac,
of course.

Thanks!

 > 0compile -v autocompile http://ryppl.github.com/feeds/boost/system-dev.xml
INFO:root:Loading configuration from C:\Users\dave\AppData\Roaming\0install.net\injector\global
INFO:root:Scheduling new task: recursive_build
INFO:root:Scheduling new task: quitter
INFO:root:Scheduling new task: solve_with_downloads
INFO:root:Task 'recursive_build' stopping and waiting for '(<Blocker:solve_with_downloads>,)'
INFO:root:Task 'quitter' stopping and waiting for '(<Blocker:recursive_build>,)'
INFO:root:No compile <command> in http://ryppl.github.com/feeds/boost/system-dev.xml
INFO:root:No compile <command> in http://ryppl.github.com/feeds/boost/system-dev.xml
INFO:root:Scheduling new task: download_done_stats
INFO:root:Scheduling new task: download
INFO:root:Scheduling new task: fetch_feed http://ryppl.github.com/feeds/boost/system-dev.xml
INFO:root:Scheduling new task: monitor feed downloads for http://ryppl.github.com/feeds/b

Testing fixes to python ZI on windows

Dave Abrahams — 2012-05-11T13:46:50

Suppose Thomas makes an upstream fix to zeroinstall, or I need to test a
fix to Python ZI on the windows version?  Is there a reasonable
procedure?

Thanks,