If you translate the entire simple-method into a groovy script (perhaps using an FTL file) and then compile/cache/run that groovy script you can avoid these overhead problems... and also trim the size of the code to execute simple-methods by a LOT (ie all those Java objects per simple-method operation would go away).
As an example the template in Moqui that does just this is available here: https://github.com/jonesde/moqui/blob/master/framework/template/XmlActions.groovy.ftl One nice thing about this is that everything is groovy by default so you can inline it anywhere. For debugging whenever there is an error just log the generated groovy script (with lines numbers) along with the error message so you can easily see where the problem happened and why. -David Adrian Crum wrote: > Replacing FSE with Groovy is a bad idea. Adam and I optimized FSE so > that it is very lightweight and fast. I also optimized the UEL > integration so there is very little overhead in the evaluation process. > Switching everything to Groovy will slow things down and increase memory > usage. Also keep in mind that Groovy creates a class for every script, > so we will run out of permgen space again. > > I think a wiser strategy would be to make mini-lang as feature complete > as possible, and include a from-script attribute for any feature gaps. > In other words, use from-script as a last resort - because it is costly. > > -Adrian > > On 3/6/2012 8:53 AM, Jacopo Cappellato wrote: >> On Mar 6, 2012, at 9:32 AM, Adrian Crum wrote: >> >>> I don't understand what you mean by supporting a limited number of >>> types. Currently, mini-lang supports any type - thanks to the >>> conversion framework. >> The conversion framework is fine; I was thinking that we could >> implicitly (by default) treat in Minilang all the numbers as >> BigDecimals, all the strings as GStrings/Expandable Strings; where >> special conversions are required than the type can be specified. >> >>> I like the idea of changing the from-field attribute to from. I would >>> like to see a from-script attribute added: >>> >>> <set field="field4" from-script="groovy: parameters.inputField1 + >>> 10"/><!-- Use Groovy --> >>> >> and why not: >> >> <set field="field4" from="parameters.inputField1"/><!-- Use Groovy >> internally: refactor OFBiz custom code to delegate on Groovy the >> evaluation of simple assignments; this could potentially replace >> FlexibleStringExpander related code --> >> <set field="field4" from="groovy: parameters.inputField1 + 10"/><!-- >> Use Groovy explicitly to evaluate the expression (use the same "from" >> attribute instead of a separate "from-script")--> >> <set field="field4" from="parameters.inputField1 + 10"/><!-- Use >> Groovy (by default, configurable) to evaluate the expression--> >> <set field="field4" from="beanshell: parameters.inputField1 + >> 10"/><!-- Use Beanshell to evaluate the expression--> >> >> ? >> >>> Then we can remove script support from expressions, which will >>> eliminate ugly hacks like: >>> >>> <set field="field4" value="${groovy: parameters.inputField1 + 10}"/> >>> >> +1 >> >> Jacopo >> >>> -Adrian >>> >>> >>> On 3/6/2012 7:31 AM, Jacopo Cappellato wrote: >>>> I am a big fan of Minilang too. >>>> The "evolution" strategy that I would like to see implemented for >>>> Minilang is actually the same one I would liketo see applied to >>>> OFBiz framework in general: review the current usage of the tool, >>>> fix existing usage for consistency (upgrade old code to use newer >>>> mechanisms offered by the tool), get rid of unused or old mechanisms >>>> in the attempt to slim down the size of the framework code, >>>> unify/simplify mechanisms based on lesson learned; all of this could >>>> be useful even to prepare the future migration to a different tool >>>> (e.g. Groovy). >>>> >>>> I know that it is very vague and doesn't add much to this thread but >>>> I like the approach suggested by Adrian. >>>> In my opinion, a good way to define a new version of the "set" >>>> operation could be that of analyzing how we are currently using the >>>> operation in OFBiz: as a starting point we could start by searching >>>> all occurrences of "<set " string in OFBiz, then review them and see >>>> different patterns; discuss and define the few ones that we like >>>> more, convert all code to use them consistently, then (or in the >>>> same phase) define the new element to better implement the patterns >>>> that we like. >>>> >>>> And now I am switching to the "brainstorming" mode :-) >>>> >>>> Kind regards, >>>> >>>> Jacopo >>>> >>>> ======================== >>>> <brainstorming> >>>> I would like to have a "set" operation that implements some of the >>>> ideas of the "configure by exception" concept. >>>> As regards the type supported, but pending the review of existing >>>> usage, we may consider to only support these: >>>> >>>> * Object >>>> * List >>>> * Map >>>> * BigDecimal/BigInteger (all numbers in Minilang should be treated >>>> as BigDecimal; no support for Integer, Float etc...) >>>> * String (expander i.e. the equivalent of GString in Groovy) >>>> * a date object >>>> >>>> Then we could get rid of the "from-field" attribute and replace it >>>> with a "from" attribute that can take as input a single field (as it >>>> is now) or an expression; some examples (all the following are >>>> evaluated using Groovy except where a different language is >>>> specified i.e. default scripting language): >>>> >>>> <set field="field1" from="parameters.inputField1"/> // field1 will >>>> have the same type of inputField1 >>>> <set field="field2" from="parameters.inputField1 + >>>> parameters.inputField2"/> // if inputField1 and inputField2 are >>>> numbers then field2 will be the BigDecimal sum of the two >>>> <set field="field3" from="parameters.inputField1 * 10"/> >>>> <set field="field4" from="script:bsh parameters.inputField1 + >>>> 10"/> // use Beanshell >>>> <set field="field5" from="parameters.inputField1" >>>> type="BigDecimal"/> // if inputField1 is a string representation >>>> of a number we can convert with the explicit definition of the type >>>> >>>> For the constant values (I am not sure if it is a good idea, but for >>>> now I will throw it out): >>>> >>>> <set field="stringField" value="This is a string"/> >>>> <set field="stringField" value="This is a string with a ${variable}"/> >>>> // the following two are equivalent >>>> <set field="bigDecimalField" value="100"/> // the system attempt >>>> to parse "100" as a number first (BigDecimal) and then as a string >>>> <set field="bigDecimalField" value="100" type="BigDecimal"/> >>>> <set field="stringField" value="100" type="String"/> // treat the >>>> field as a string >>>> >>>> </brainstorming> >>>> On Mar 5, 2012, at 9:07 PM, Adrian Crum wrote: >>>> >>>>> I am not one of those people. I use mini-lang almost exclusively. >>>>> >>>>> -Adrian >>>>> >>>>> On 3/5/2012 7:46 PM, Anil Patel wrote: >>>>>> Adrian, >>>>>> Thanks for starting this thread. >>>>>> >>>>>> While we all love mini-lang, I am wondering if we should really >>>>>> ask ourselves if we really want to overhaul mini-lang or should we >>>>>> consider alternates. From what I know, Not many people like to >>>>>> build application using mini lang. Many end up using Java or Groovy. >>>>>> >>>>>> Thanks and Regards >>>>>> Anil Patel >>>>>> HotWax Media Inc >>>>>> >>>>>> On Mar 5, 2012, at 9:47 AM, Adrian Crum wrote: >>>>>> >>>>>>> Mini-language has evolved a lot over the years. Most of the >>>>>>> development has occurred on an as-needed basis, so there is no >>>>>>> clear design or implementation - things just get tacked on over >>>>>>> time. >>>>>>> >>>>>>> A recent discussion has opened up the possibility to rework the >>>>>>> mini-language<set> element. From my perspective, that task is >>>>>>> long overdue. >>>>>>> >>>>>>> Also, the schemas are out of date, and they are unnecessarily >>>>>>> complicated. So, those need a thorough going over. >>>>>>> >>>>>>> While we are at it, why don't we create a draft design document >>>>>>> based on the current implementation, and then use it to look for >>>>>>> other ways mini-language can be improved? We can all offer >>>>>>> suggestions and comments, agree on a final design, finalize the >>>>>>> draft, and then implement it in code. The design document then >>>>>>> becomes the developer's reference. >>>>>>> >>>>>>> What do you think? >>>>>>> >>>>>>> -Adrian >>>>>>>
