After spending some time working with Flex 4 on a real project, I can’t say I’m happy with this product. My list of complains is based on moderate sized project, 5 months, 4 developers, WebService-only back-end.
1. Flash Builder 4 has lots of issues – more than Flex Builder 3. AS of today it has lots of bugs:
-it doesn’t release resources;
-debugger does not allow to run several instances and consumes a lot of resources;
-debugger intercepts all network calls and the debugging proxy is terribly slow and inefficient;
-code navigation between several projects is absolutely impossible;
-compiler is unpredictable and slow;
-it periodically tries to do incremental compilation, even if you explicitly tell it not to do so — and you have to disable incremental compilation, because it’s slower than full rebuild of all related projects with Maven.
-Help system is not well designed — Adobe moved help to a separate AIRr application, in-place contextual help is non-existent; search always shows one link — “yes, there are probably some topics for your query — open separate application to search again”
2. Flex compiler is broken
It’s slow. It’s slower than any previous version I’ve used. And it’s buggy. Periodically it shows some cryptic messages like “there is no x or y or width or height properties on UIComponent”. It never points to exact source of problem — and I doubt it exists. Because sometimes fix is to set “-keep-generated-actionscript” flag on. And sometimes off. And if it doesn’t help then the fix is to reorder attributes in MXML tags. Just insane. And I can’t find the exact root of these messages, the only thing I can say for sure that they are caused by new MXML skins — no skins, no errors.
3. The new Spark library is over-engineered.
I worked on some Java Swing projects some time ago. It seems that Adobe takes the Swing route, however Flex becomes popular just because of the opposite — it was way far simpler than Swing. Flex ideology was “keep simple things simple, make complex things possible”. Now it turns to be the Swing ideology: framework usage is irrelevant, the only relevant thing is framework per se, and it should be perfect.
I can’t do a lot of things with Spark I was able to do with MX. Icons in buttons? Develop descendant component of Button, define new parts and develop separate skin. ComboBox where items in drop-down are large than input field? Develop own skin — just copy-paste and tweak a bit. Paddings? What paddings? It’s not a css attribute any longer, extend and develop own skin. Left/right/top/bottom? You can’t customize this via css, or develop your skin if you truly need this option.
But architecturally Spark is beautiful, no doubt. Astronauticaly beautifull architecture, if Joel Spolsky permits me to use this term.
4. New Spark library is incomplete.
Yes, this statement doesn’t contradict with . Being overdesigned it’s missing a lot of functionality from MX: no Calendar, no DateInput, no ColorPicker, no PopUpButton etc. No clone for MX SWFLoader/Image. There is no complex containers at all (like Accordion and TabNavigator) Needless to say, there is no advanced controls like DataGrid and the supplied “proxies” for ItemRenderers confirm that there are will be none any time soon.
5. New states engine encourages unmaintainable code
Previously we have states with explicit overrides. All data related to override was co-located, so it was easy to see what and when changes. There were inheritance of states, and it was sufficient. There were no groups, but when you need state groups — then 99.9% of times you in fact have to extract part of component as sub-component, and coordinate isolated state changes there. But now the great property.inStateFoo=”xyz” notation comes.
Now you have to scan over tens or hundreds of lines of code and check all and every includeIn / excludeFrom attribute and decompose them back to the very same old “overrides”. In short, when you are debugging states you have to do job of compiler but without the help of compiler. And the existence of state groups multiples the complexity by the factor of ten. It’s a Ruby-way productivity — it’s easy to develop a sketch / prototype, it’s a nightmare to polish this sketch to be a real product and it’s absolutely impossible to support it. But sure, you can start quickly, who cares about the rest…
6. Vectors (Flash-10 related, not just Flex 4)
Vector is a canonical example of a half-pregnant woman. Hey, developers! Use new features – Vectors and Generics (or this is templates??? we are not sure on our own). Yep, it’s cool, Vector is a parametrized type. Though, there is no type covariance/contr-variance and no type bounds in syntax. So you have to cast data Vector whenever it’s declared as parameter or as a return value. But who cares?
7. Text engine (Flash-10 related, not Flex 4 only)
It’s great that “Gordon works on text”, and new engine is indeed superior. But may we have both Spark and MX on same engine by default without the dual way to embed fonts and dual size of embedded fonts?
Am I missing something?