Felix Schwarz Diplom-Informatiker
Software-Entwicklung und Beratung

Free Software is no silver bullet for niche markets

Many people think of solid application which causes not too much trouble when they talk about Free Software. Most Free Software is given away for free, everyone can extend it as (s)he likes (presuming the necessary knowlegde) and there is no marketing department which urges for premature 'final' releases. Administrators often like that Free Software does not require any license management (if you just use it) and bug reports are openly shared because of the abundance of company politics which may try to maintain a zero-defect illusion even if the product is buggy like hell.

But the article's head line is 'Free Software is no silver bullet for niche markets' so you probably already expect that I will give you some counter-examples to the cliches above.

Pipe dreams meet real live

So today's example is Lx-Office (Wikipedia, sorry no English version available). Lx-Office is a web-based ERP and CRM system written in Perl and PHP and distributed under the GNU General Public License. It includes a module for financial accounting in companies. The ERP/accounting module of Lx-Office originated in SQL-Ledger (Wikipedia). After the fork from SQL-Ledger in 2003, Lx-Office was modified to meet German regulatory requirements.

In Germany, the majority of all tax advisors (approximately 80% according to Financial Times Germany) uses software produced by a German cooperative named DATEV (Wikipedia). This software is really the heart of a tax advisor's services, different versions support almost everything from financial accounting and tax calculation to payroll accounting.

So you can imagine that for a serious accounting software, data exchange with tax advisors is really a must-have. Though DATEV does only offer properietary software (which only runs on MS Windows), they sell a SDK which contains the complete specification of the KNE format ("Postversandformat"). With that format you can share accounting information with your tax advisor (e.g. the routine accounting is done in-house but the tax advisor should assemble the balance sheet at the end of the year).

The KNE format is an 'old-style' binary file format with a long track record (if you include the predecessor, the OBE format which was introduced in the late '80s). When I wrote libkne, a pure Python library to create and parse these binary KNE files, I tested my implementation against several real-world implementations. However, when I used the DATEV export using Lx-Office's demo server I was very surprised to see that the exported files where not valid according to the specification!

Export bug known for several years

It turned out that the problem is caused by different date string formatting. The export module DATEV.pm transforms a string which is presumed to contain a date ('25.12.2008', which is 2008-12-25 in ISO notation) into the KNE binary date format ('251208' - yes, the KNE format is not y2k-safe). As soon you use a different date format (like '12/25/2008'), the export just puts in the date string '12/25/2008', ignoring the size restrictions (6 bytes) and the format specification ('DDMMYY') completely.

What really puzzled me was that this bug is known for 2 1/2 years and nothing happened so far. In June 2006 there was a post in the user forum which explained the problem and even spotted the bug in the source code. Only thirty minutes later, another forum reader filed a bug report with a reference to the post. Six months later, a developer requested more information (which looks stupid to me, all necessary information was given) from the submitter but no further action was taken until now (December 2008). That means that the DATEV export of Lx-Office is non-functional for 2 1/2 years if you don't use one specific date format!

Looking at the source

As always, we can have a look at the source to reveal the issue:

# Code taken from SL/DATEV.pm (r3482)
# https://lx-office.linet-services.de/svn/trunk/unstable/SL/DATEV.pm
# Licensed under the GPL v2+

sub datetofour {
  $main::lxdebug->enter_sub();

  my ($date, $six) = @_;

  ($day, $month, $year) = split(/\./, $date);

  if ($day =~ /^0/) {
    $day = substr($day, 1, 1);
  }
  if (length($month) < 2) {
    $month = "0" . $month;
  }
  if (length($year) > 2) {
    $year = substr($year, -2, 2);
  }

  if ($six) {
    $date = $day . $month . $year;
  } else {
    $date = $day . $month;
  }

  $main::lxdebug->leave_sub();

  return $date;
}

So the datetofour function gets a date string $date and a boolean flag $six and returns the KNE date string. The problem is that the function assumes a certain date format in line 10 in the snippet (line 481 in original file as of r3482): 'split(/\./, $date)' But this date formatting can be configured by the user so the function should query the configuration for the format to use.

Conclusions

I presented you one example of a Free Software application which has a serious known bug that was not fixed for more that 2 1/2 years (though it is probably only a matter of one or two lines). The project itself is backed by two small companies and the software is used by quite a lot of people (Sourceforge counted 4468 downloads for Lx-Office ERP beta1 between 2008-08-12 and 2008-12-25). The project follows a traditional 'open source' development approach (public bug tracker and source control repository, community support through web forums) and uses a well-known license (GPL 2+). So this is clearly an example of a 'real' Free Software project.

Nevertheless the project failed to fix a simple bug so that users can not use functionality which is quite important in many enterprise use cases if you consider the extensive German regulatory policies. Obviously this functionality is not important enough for most users of Lx-Office. Probably (speculation!) many of them are in the 'low budget' segment which does not regularly exchange financial data with their tax advisors.

Admittedly financial accounting is really a niche use case (highly dependend on the country you're operating in, laws and policies change on a regular basis) so the user and developer number is quite small (compared to other software like Linux, Mozilla Firefox or OpenOffice). But it still gives evidence to my initial thesis that Free Software is not necessarily a solution for you. While Free Software in very common use cases can excel their proprietary competitors (e.g. Apache HTTP), you may find that you should evaluate Free Software in niche markets very carefully.

Brief remark: 'Batteries included' is the right approach for class libraries

While I described what code caused the bug on a technical level, the real (technical) problem for me is found below: In Perl 5 there is no built-in data type to represent dates. So the Lx-Office developers had to fall back on using strings to pass dates around. If Perl shipped some included module for date representation and manipulation, probably the bug would have never occured because things like date formatting would only influence the visual presentation.

So I think that on a class library level Python's approach of batteries included is the right one. A decent(tm) programming language needs to ship modules for date manipulation (even a less-than-optimal API like in Python is better than a completely missing one). Just having add-on modules for that is not sufficient because not all developers use them and it's hard to use multiple independently developed third-party modules with each other because they probably will use incompatible date representations...