Bug#522082: tla: file descriptor leak, causes panic when removing temporary directory

Package: tla
Version: 1.3.5+dfsg-14
Tags: patch

When computing a changeset (e.g. for the "tla changes --diffs" command), tla
creates a temporary directory with a name that begins in ,,what-changed.
This directory is cleaned up at the end of the operation.

Symbolic links are compared by writing the target of the link into a
text file under that directory, then relying on usual "diff" behaviour.

Unfortunately, tla doesn't explicitly close this text file. As a result:
(1) it uses more file descriptors than necessary;
(2) on some filesystems (e.g., OpenAFS 1.4.8 and newer) tla dies with
an I/O panic as the rmdir() operation on the parent directory fails with 
ENOTEMPTY, even though the unlink() of the file itself succeeded, because 
the file is still open.

The attached patch has been tested and gets rid of symptom (2). I expect
it to help with (1) as well. The variables that hold the file descriptor
numbers go out of scope shortly afterwards.

#! /bin/sh /usr/share/dpatch/dpatch-run
## 07-changeset-fd-leak.dpatch by Sergio Gelato <Sergio.Gelato <at> astro.su.se>
##
## All lines beginning with `## DP:' are a description of the patch.
## DP: Remember to close file descriptors before they go out of scope.

 <at> DPATCH <at> 

please look into gittorrent

Please look into gittorrent.  While there are still important (at least
to my mind) gaps between git and Arch (things missing from the former
that are at least pointed out by the latter) nevertheless gittorrent
looks initially like an important development that helps narrow the gap.
I wanted to go someplace similar with GNU Arch.

Arch does better than git in taxonimizing versioned objects and in its
management of coding history, branching, merging, etc.... but Arch
shares with gittorrent this idea of distributed, decentralized revision
control -- free software source code should be just sort of "floating"
on a meta-net, on the Internet, over a P2P layer -- just so.  This is a
political goal because of the "decentralization" part.

We also have a lot of work to do on the economics of the emerging
ecosystem of free software source code and because economics "wants" to
be "transactional" -- and because of the nature of the natural unit of a
"transaction" in software source code development and support -- the
economics of how free software can be a career and the formality imposed
by a global-scale, distributed, decentralized revision control system
are closely intertwined.  Issuing a "commit" command should be an
economically significant act -- a "transaction" in both senses of the
word.

But that's a larger, future topic, for now.   For now: please do look
into gittorrent and share your impressions.  Have they actually made
progress on distributed, decentralized revision control?  Are their
politics in order?

Thanks,
-t

Inclusion of Bzr into the GNU system

I've been sitting on this for a long time, but it's been bothering me a 
bit and I need to get some others' perspectives on it.

I'm a fan of GNU Arch.  Seriously.  I'm not a major software author at 
all, but I do have need of a revision control system, and of all that 
I've used, GNU Arch is the one that I really feel fits how I work.  The 
fact that it is distributed, that it uses forward patching, that it has 
a sane, usable interface and more make it ideally suited to my needs.  
Most of all, it's a GNU project, and thus I am ensured that in using it 
I retain my freedom.

The fact that GNU Arch has a smaller user base than some revision 
systems has always been a bit of a downer to me, but has not deterred my 
use of it in my own projects.  When Bazaar forked, then rewrote, GNU 
Arch, I was fairly unimpressed, because potential users of Arch were 
then pulled over to Bzr.  I can't fault either the Bazaar team nor the 
users; this is a freedom protected by free software.  I would have 
preferred to see those same efforts make there way into GNU Arch (though 
I have read of the reasons why they weren't), but what really galls me 
is the means of developing Bazaar and the subsequent product.

Bazaar is written in Python.  I've used Python, and though I personally 
don't prefer it (I'm a Guile user), I see its merits and why so many 
programmers are drawn to it.  But Python, while ostensibly free 
software, has a rather weak license and in the past its community has 
shown a bit of hostility toward maintaining the Four Freedoms.  This 
greatly troubles me for a project that has been accepted as part of the 
GNU system.  I would like to have seen the deliberations over accepting 
Bazaar as a GNU project and know whether this issue was brought up.

botched invariant pest

Greetings,

In recent days I am experiencing the same error message again
and again, a message which does not reveal anything to me:

* looking for juengling <at> pdx.edu--2008/lush--soc--1.4--patch-176 to compare with
* comparing to juengling <at> pdx.edu--2008/lush--soc--1.4--patch-176
/u/juenglin/downloads/tla-1.3.5/src/tla/libarch/invent.c:982:botched 
invariant
     *a < *b
PANIC: exiting on botched invariant

Whenever I run into this, the project tree becomes unusable and
I need to start a new one, find those modified files and copy
them over to the new project tree by hand.

These error message are haunting me for a couple of days now
and have not been able to make out what might trigger them to
occur. I am working on my files, do a 'tla changes' once in a
while, and suddenly I see this error.

Any help will be much appreciated.
Ralf

GNU Arch intro in russian

	FWIW, I've put a short introduction into the GNU Arch in Russian
	on the Web:

http://theory.asu.ru/~ivan/doc/tla/index.html.ru

	The whole idea behind these bits was to make new GNU Arch users
	able to do some basic work with it, such as: registering archive
	locations and retrieving working copies out of the Arch
	archives.  It doesn't reveal much more beside of that.

	(Although I still have plans for writing a more comprehensive
	comparison of the GNU Arch to the other DVCS of today, both in
	English and Russian.)

revc

Hi,
In case someone cares, I've been able to find somewhere the last
revision of revc made by Thomas Lord. sha1sum and md5sum are the same
as the ones given on the website.
You'll find the file here: http://dl.kodros.fr/revc.0.0x2.tar.gz
Regards,
Laurent.

[Help-gnu-arch] Receiving error with tla command under OSX Leopard 10.5.1

I upgraded my MacBook Pro to OS X Leopard 10.5.1.

When I issued a tla get <project> <local directory> I received the  
following error:

keaton:~/arch/pandora$tla get sql--pandora--0 sql-pandora
Error during call to `vu_access' for /home/Arch/revlib/=greedy (Input/ 
output error)
PANIC: I/O error

This was working just fine under OS X Tiger (10.4.x).

I recompiled tla under Leopard with the updated XTools and tried  
again, but received the same results.

Any ideas?

Thanks,

Keaton

TLA: patches for neon >= 0.25.2 and others

When built with neon >= 0.25.2, tla-1.3.5 crashes. Some distributions
(eg., Fedora) prefer to build packages against the system's libraries
instead of libraries embedded in a upstream tarball -- in this case,
libneon.The problem was reported in both Debian and Fedora:

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=402952
https://bugzilla.redhat.com/show_bug.cgi?id=327111

The tla package in Fedora uses the following patch to fix this issue:

diff -urNp tla-1.3.5.orig/src/tla/libarch/pfs-dav.c
tla-1.3.5/src/tla/libarch/pfs-dav.c
--- tla-1.3.5.orig/src/tla/libarch/pfs-dav.c	2007-12-13 22:51:47.000000000 +0530
+++ tla-1.3.5/src/tla/libarch/pfs-dav.c	2007-12-13 22:54:36.000000000 +0530
 <at>  <at>  -128,7 +128,7  <at>  <at>  static t_uchar * abs_path (const t_uchar
                            const t_uchar * path);
 static t_uchar * dirfold (t_uchar *dir);
 static void results (void * userdata,
-                     const char * uri,
+                     const ne_uri * uri,
                      const ne_prop_result_set * set);

 
 <at>  <at>  -988,22 +988,23  <at>  <at>  dirfold (t_uchar *dir)

 static void
 results (void * userdata,
-         const char * uri,
+         const ne_uri * uri,
          const ne_prop_result_set * set)

dear colleagues

Well, perfect end to a fucked up decade......

Really exciting night!   Someone came pounding on my door to call me 
outside.   Turns out, someone has totaled my car while it was parked on 
the street for the night.

Why is this exciting?   Because I can't walk much at the moment and that 
car was vital to day to day survival.   I wonder how I'll manage to get 
groceries tomorrow.

Wait, it gets better.

Being poor, I'm fully paid up and current on lowest available rate 
insurance.   So, I called the claims number tonight and, an experience 
familiar from the open source world, the number is disconnected.

Watch us drown.

If anyone would like to help either please do so or pass this message 
along to someone sensitive who can.

Time is of the essence, if anyone happens to give a damn.

-t

[Help-gnu-arch] Help composing a sample session

I'm trying to write a sample Arch session for a survey paper I'm
working on, and failing.

Here's my sequence of commands;

  # Clear all arch state
  rm -fr {arch} {archives} ~/.arch-params/ .arch-ids foo.txt ++log.example--trunk--0.1--esr <at> thyrsus.com--2007-example

  tla my-id "Eric S. Raymond <esr <at> thyrsus.com>"
  mkdir {archives}
  tla make-archive esr <at> thyrsus.com--2007-example {archives}/2007-example
  tla my-default-archive esr <at> thyrsus.com--2007-example
  tla archives
  tla whereis-archive esr <at> thyrsus.com--2007-example
  tla archive-setup example--trunk--0.1
  tla categories
  tla branches example
  tla versions example--trunk
  tla init-tree example--trunk--0.1
  echo "This is a one-liner file." >foo.txt
  tla add foo.txt
  tla make-log

The last command echoes this:

++log.example--trunk--0.1--esr <at> thyrsus.com--2007-example

I then edit the log file, 

The problem cones when I try to commit:

Announcing Flower 0.5

<div>
<div class="moz-text-flowed" lang="x-western">Dear
Colleague,
<br><br>
I am pleased to announce the availability of Flower 0.5.&nbsp;&nbsp; Today I have
released version 0.5 of Flower under the Open Software License version
3.0 (an Open Source license).&nbsp;&nbsp; You can find it at my web site, <a class="moz-txt-link-freetext" href="http://www.basiscraft.com">http://www.basiscraft.com</a>
<br><br>
Flower is a new kind of <span class="moz-txt-tag">/</span>user programmable web service<span class="moz-txt-tag">/</span>, especially well suited for
applications which process, store, and query XML data sets. Clients of
a flower web service interactively modify and extend the code the
server runs. This is is the ordinary way to build new flower
applications. Flower is a true <span class="moz-txt-tag">/</span>web operating system<span class="moz-txt-tag">/</span> in the sense that it forms a
self-contained, web-addressable computing environment.
<br><br>
Regards,
<br>
Thomas&nbsp; ("-t") Lord
<br><a class="moz-txt-link-abbreviated" href="mailto:lord <at> emf.net">lord <at> emf.net</a>
<br>
510-825-7915
<br>
Berkeley California
<br><br><br><br>
</div>
</div>