Hi Greg, I run into merge conflict during merge release branch to develop into file which you have changed lately. Could you please verify on develop if I didn't remove anything, if I resolve conflict correctly. [1]
[1] https://github.com/apache/royale-compiler/blob/develop/compiler-jx/src/main/java/org/apache/royale/compiler/internal/codegen/js/jx/MemberAccessEmitter.java Thanks, Piotr śr., 2 paź 2019 o 11:06 Harbs <harbs.li...@gmail.com> napisał(a): > OK. > > I’ll test memory with and without your changes and let you know the > differences. :-) > > Harbs > > > On Oct 2, 2019, at 11:29 AM, Greg Dove <greg.d...@gmail.com> wrote: > > > > @harbs > > > > FYI in addition to some other stuff, I am close to pushing my updates to > > XML. This should be in the next hour or so. > > > > I kept the QName caching pretty close to how you had it, with only some > > minor changes. > > I tried to do some extra memory optimization, and in theory it should > > provide better results, but to be honest I don't have a good way to > measure > > this in the browser. I tried the Chrome performance.memory extensions > but I > > don't have much confidence in that given how much it varies between > reloads > > without changing anything else. The extra code changes I made were to > make > > the '_nodeKind' strings into String object references, so they only refer > > to a single reference to a string instead of multiple copies of > primitives. > > That change is isolated to a single commit so can easily be reversed if > > there is something not good about it... but all my local tests continue > to > > pass. I will get the new tests into RoyaleUnit in the coming days. > > > > > > > > > > On Thu, Sep 5, 2019 at 7:39 AM Greg Dove <greg.d...@gmail.com> wrote: > > > >> Yeah, I saw that ;) Don't worry, I am aware of it. > >> > >> My first goal is to make sure it works like it should, because that > comes > >> first, and then to optimize. I'll check the memory side of things and > make > >> sure it's at least the same as before. If you can point me to some > >> publicly accessible large test cases that would be really helpful. I > will > >> work through that before I push anything. > >> > >> On Thu, Sep 5, 2019 at 7:26 AM Harbs <harbs.li...@gmail.com> wrote: > >> > >>> Heads up: > >>> > >>> I did some (on first blush) odd things in XML related to QNames. QNames > >>> are pooled and many XML properties are not initialized by default. The > >>> reason I did this was it avoided many MB of memory waste for complex > XML. > >>> Please don’t mess that up. > >>> > >>> Thanks, > >>> Harbs > >>> > >>>> On Sep 4, 2019, at 1:02 PM, Greg Dove <greg.d...@gmail.com> wrote: > >>>> > >>>> This is particularly for Harbs and Yishay, as I think you are both (or > >>> both > >>>> have been) using XML quite a bit. I have quite a few fixes coming. > All > >>>> with tests that match on swf and js. > >>>> > >>>> I am currently working to demonstrate proof of concept to a > prospective > >>>> client for migration of a Flex app. The app makes extensive use of e4x > >>> and > >>>> uses a bunch of features that I expect had not received attention > >>>> previously, because they were originally either not working with the > >>>> codebase I am porting, or i think some even caused errors in the > >>> javascript > >>>> output. > >>>> > >>>> So I have spent the last several days almost full time figuring things > >>> out > >>>> and working on fixes, between the compiler and emulation classes. > >>>> All the previous XML tests continue to pass, but I have many more unit > >>>> tests and fixes lined up for the following: > >>>> > >>>> namespace directives > >>>> default xml namespace > >>>> use namespace (multiple) > >>>> > >>>> a number of fixes for xml filtering, including: > >>>> -'this' resolves correctly in filters that include external references > >>> from > >>>> the fitler expression to the 'this' scope > >>>> -handles alternate ordering of comparisons between XML 'getters' and > >>>> literals > >>>> e.g. something.(name() = "cat") or something.("cat" = name()) (these > >>> are > >>>> the same) > >>>> -it (will) now handle XML e4x references in nested function calls > inside > >>>> the filter, e.g. things like: > >>>> e.g. > >>>> var people:XML = <people> > >>>> <person> > >>>> <name>Bob</name> > >>>> <age>32</age> > >>>> </person> > >>>> <person> > >>>> <name>Joe</name> > >>>> <age>46</age> > >>>> </person> > >>>> </people>; > >>>> var findJoeByAge:Function = function (i:int):Boolean { > >>>> return i > 40; > >>>> }; > >>>> people.person.(findJoeByAge(parseInt(age))).name > >>>> > >>>> > >>>> I have lots more granular tests in QName, Namespace, and XML with > >>> tuning to > >>>> improve reliability. > >>>> toXMLString XML node output also matches flash more correctly in what > I > >>>> have coming. > >>>> > >>>> One thing that I am trying to figure out, which I would appreciate > >>> input on > >>>> if someone has an answer: > >>>> For the example: > >>>> > >>>> var feed:XML = new XML( > >>>> '<feed xmlns:atom="http://www.w3.org/2005/Atom" > >>>> xmlns:m="nothing">\n' + > >>>> ' <link rel="no_namespace" > >>>> href="blah/12321/domain/"/>\n' + > >>>> '</feed>\n'); > >>>> var atomSpace:Namespace = new Namespace('http://www.w3.org/2005/Atom' > ); > >>>> > >>>> Can anyone explain why this (in SWF, as a reference implementation): > >>>> trace(feed.atomSpace::link.length()) > >>>> trace(feed.atomSpace::link.toXMLString()) > >>>> //output: > >>>> 0 > >>>> {empty string} > >>>> is different to: > >>>> trace(feed.child(new QName(atomSpace,'link')).length()) > >>>> trace(feed.child(new QName(atomSpace,'link')).toXMLString()) > >>>> //output: > >>>> 1 > >>>> <link rel="no_namespace" href="blah/12321/domain/" xmlns:atom=" > >>>> http://www.w3.org/2005/Atom" xmlns:m="nothing"/> > >>>> > >>>> I had assumed the above cases would be the same, but the second one is > >>>> behaving as if it has the default namespace included with the > specified > >>>> namespace in the QName matching (it does correctly match the namespace > >>>> specifically as well -with e.g. <atom:link nodes where the prefix atom > >>> is > >>>> bound to the uri, but also seems to include link nodes with the > default > >>>> namespace, whether or not it is declared). I can accommodate this > >>>> difference to make them behave the same, I just would like to > understand > >>>> the basis for the actual rules if anyone knows.... > >>>> > >>>> I should be in a position to push the updates this coming weekend I > >>> think. > >>> > >>> > > -- Piotr Zarzycki Patreon: *https://www.patreon.com/piotrzarzycki <https://www.patreon.com/piotrzarzycki>*