Hi Adrian, I am inclined to say that any code improvement that removes quirkiness in current code or unpredictable behaviour while at the same time helping contributors to develop better solutions deserves a +1.
However I have some concerns regarding the approach you're following. First, as we can expect that the quirkiness in current widget code is being utilized in various places in the dependent components, refactoring the widget code can lead to a multitude of issues regarding Screens and Forms in those components. I truly hope this will be as bad as I fear. Secondly, refactoring the widget code will have an impact on performance. I am sure we all hope (and some will expect it to be so) it is a positive effect, but we can't tell - at this moment in time - what it will be. Given the two concerns I would rather have seen that the refactoring was taking place in a separate development branch than in trunk. Having it done that way we would have the opportunity to compare the performance aspects of the refactored code against trunk (e.g. with Apache jMeter and such), and if favourable have another argument to tell to the users of old releases. Also while having the refactoring taking place in the development branch, we can not only isolate the aspects of my first concern (development branch issues vs trunk issues), but we will be able to better explain what the effects of the refactoring could be visavis (potential) issues of the users of old versions considering upgrading. Best regards, Pierre Smits *ORRTIZ.COM <http://www.orrtiz.com>* Services & Solutions for Cloud- Based Manufacturing, Professional Services and Retail & Trade http://www.orrtiz.com On Sat, Jan 17, 2015 at 6:46 PM, Adrian Crum < adrian.c...@sandglass-software.com> wrote: > Some time ago I started working on the screen widget thread safety. There > were many places in code where widget models were being modified during > rendering - resulting in unpredictable behavior, and in some cases it > resulted in users having access to data they shouldn't be able to see. > > While doing that work, I was overwhelmed by the quantity of source code. > The screen widget library was built using a lot of copy-and-paste - instead > of extracting and reusing common things. Scott started working on reusing > widget code, but that was just a small beginning. > > In a recent commit, I continued his work and made some more things > reusable. > > Next, I would like to reorganize the source code folder structure. Here is > what I have pictured: > > org/ofbiz/widget > artifact (Artifact Info classes) > cache (Widget cache classes) > model (Widget models) > renderer (Widget renderers) > macro > html > xml > > I think the simplified folder structure makes more sense and it will make > it easier to locate classes. > > After that, I would like to add error checking code to the widget models - > similar to what I did in Mini-Language. Right now, errors in widget XML are > (sometimes) logged and widget parsing continues. If a developer does > something wrong, they will not know it unless they check the logs. I would > like to change the behavior so widget XML errors throw an exception with a > detailed error message that includes the XML file name and line number > where the error occurred. I believe this will benefit developers by making > it clear when they have done something wrong. > > Finally, I would like to extract list functionality from the form widget > and create a new grid widget. So, instead of a form widget representing a > single data entry form OR a list, it will ONLY represent a single form. If > you want a list, you use the grid widget. Initially, this change will be > backwards-compatible - the XML parser will accept a <form> element for both > types and it will create the correct model based on the type attribute. > > Overall, my goal is to make screen widgets more developer-friendly, and > also to make it easier to innovate in the screen widget component. > > After all of this work is completed, I would like to backport it to the > R14 branch. > > Comments are welcome. > > > -- > Adrian Crum > Sandglass Software > www.sandglass-software.com >
