This post is meant to be a helpful reference for developers who are either just beginning OpenERP development or who are considering beginning OpenERP development. It’s also a fairly thorough catalog of my gripes with it. Hopefully this will help save someone somewhere some pain. I start with a summarized list of its pros and cons, followed by a more thorough explanation of each of the cons, and finish up with a list of tips that I personally would have found useful back when I was getting started.
I’ve been working with OpenERP v7 since December 2013. (Version 8 has since been released as “Odoo.”) In the past 9 months I’ve gone from being a complete newbie to being familiar and competent with every aspect of this ERP framework. I have no basis for comparing it to other ERP solutions, as it’s the first I’ve ever worked with, but I can make a decent comparison between it and other Python-based projects.
Before I launch into describing the pain points in more detail, I should first acknowledge that OpenERP began in 2005 as TinyERP. Had I myself begun an ERP system in 2005, I might not have done much better. The ecosystem was not the same back then, and a lot of my gripes likely are the result of the hindsight that the original developers didn’t have. That said, there are a lot of things about OpenERP that tripped me up and that still trip me up from time to time. If you’re just getting started with OpenERP, learn a little from my experience and read on. (And if you have the chance to try Tryton instead, give it some consideration. From the outside, it appears to be a bit more Pythonic than OpenERP.)
The OpenERP developers seem to have built custom systems for everything. The client whose OpenERP system I was building on had advertised for a “Python developer,” but I felt like almost none of my Python knowledge got leveraged because almost everything had been reinvented in one way or another. The “workflows” system could have been done with blinker, but instead they opted to implement a custom, hybrid Python/XML-based system. The templating system could have been implemented quite nicely with Mako, but instead they opted to implement yet another custom, hybrid Python/XML-based system. Speaking of which…
Which wouldn’t be so bad if I didn’t hate XML so much. But in my opinion, XML is a poor match for Python. As I understand it, XML is great for enforcing correctness and catching errors early. It’s especially great when you’re working with a static-typed language, from what I hear. And while OpenERP does catch template errors upon module installation, the error messages aren’t especially intuitive. On top of this, a bad template with literally break the entire server. If your object models doesn’t match their templates somehow, the server will lock everyone out and answer every login attempt with a stack trace.
Rather than use Python’s package system, OpenERP uses a custom package system. “Add-on paths” are defined in the server’s configuration file in order of priority, and modules within these add-on paths can import each other and override each other’s code.
It’s a fairly flexible system, but it makes tracing code to find bugs more complicated than it has to be. In a normal Python project, figuring out what code is executing is as simple as looking at the imports at the top of the file and following them all the way up. With OpenERP, you still have normal Python imports to worry about, but now you also have OpenERP’s system to worry about as well. In OpenERP’s system, module imports are defined as strings in an array in openerp.py, with their order in the array determining in which order they get imported. This in turn determines which version of, say, “res.company” your module’s “res.company” class inherits from. Because in OpenERP, class definition and inheritance are determined by magic class properties. If the name string is set, the class is being defined, which can lead to occlusion if the name is used in a previously imported module. If the inherit string is set, then the class inherits from a class belonging to the last-imported module whose inherit or name property matches.
This system is, in my opinion, needlessly complicated and unidiomatic. I could be wrong, as I’ve never written an ERP myself, but I suspect simply allowing developers to leverage their existing knowledge from, say, SQLAlchemy might have been a better design decision.
Upgrading the database after changing the codebase doesn’t always work the way you would expect it to, for example. This has led to fun situations where, for example, a module works on my dev box but not on the test box because once the module had defined a column on one of its models, and that column wasn’t removed from my dev database during an upgrade and some other package turns out to have relied on that column being in the database.
Installing a new module can likewise be treacherous. I once spent hours trying to figure out why the new menu items I had added to a module weren’t appearing, despite my settings in the security permissions CSV being correct, and despite the menu items showing up under the menu items tree in the user interface settings section of OpenERP. There were no errors whatsoever during installation, and there were no errors from the module during runtime. In the end, it turned out that one of the modules in its dependency chain was having trouble connecting to a remote server, which somehow led to the new menu items not being displayed. I understand that managing actors that are able to unilaterally alter each other’s states is a very difficult problem. I understand that mutating shared, global state is going to happen within the context of an ERP. But I can’t help but feel it could be done a whole lot better.
I know there’s some dissent about ActiveRecord’s usefulness, but it is a very common pattern at this point. OpenERP instead uses a very restricted, custom ORM that requires passing around cursor handles, user IDs, object IDs, and contexts. Because of how restricted the ORM is, very few of the efficiencies that can be realized by eschewing an ActiveRecord pattern are realized, and you get none of ActiveRecord’s idiomatic syntax.
Don’t get me wrong: there’s a lot of it out there, and the fact that there is documentation at all puts them ahead of a lot of projects. The problem is primarily that it’s not organized well. There isn’t a clean division between documentation that would primarily appeal to developers and business users. The style tends to veer toward verbose tutorials rather than plain API listings. There are holes here and there, where I’ve been forced to just dig into the source code to figure out how things work. There are different versions of the documentation for different versions of OpenERP, each with unique holes, which leads to jumping between versions in order to fill in the holes and hoping that what held true for version 5.0 still holds true for version 7.0.
And perhaps worst of all, there are the blank pages and dead links that started popping up everywhere after the developers rebranded OpenERP as Odoo. I use StartPage and IxQuick to search for OpenERP documentation because they tend to work better than the search on the OpenERP/Odoo website. But they’re not working quite as well as they used to, because oftentimes about half the results in the top five will be either to a thread on their old forum or to a page that appears to have literally just had its content removed. The latter case is particularly frustrating: the heading is still there, the page is still in the table of contents for that version of OpenERP’s documentation, but there is no longer any content.
If this hasn’t scared you away from OpenERP, here are a couple pointers that I wish someone had given me when I first got started:
Best of luck in your OpenERP/Odoo adventures! Let me know your thoughts and opinions in the comments, and feel free to ask questions. I’ll do my best to answer them.