[ovirt-devel] Custom Properties code duplication

Lior Vernia lvernia at redhat.com
Wed May 7 16:29:23 UTC 2014 


On 07/05/14 18:57, Gilad Chaplik wrote:
> ----- Original Message -----
>> From: "Lior Vernia" <lvernia at redhat.com>
>> To: "Gilad Chaplik" <gchaplik at redhat.com>
>> Cc: "Vojtech Szocs" <vszocs at redhat.com>, devel at ovirt.org
>> Sent: Wednesday, May 7, 2014 5:32:38 PM
>> Subject: Re: [ovirt-devel] Custom Properties code duplication
>>
>>
>>
>> On 07/05/14 16:02, Gilad Chaplik wrote:
>>> ----- Original Message -----
>>>> From: "Lior Vernia" <lvernia at redhat.com>
>>>> To: "Vojtech Szocs" <vszocs at redhat.com>
>>>> Cc: devel at ovirt.org
>>>> Sent: Monday, May 5, 2014 7:08:23 PM
>>>> Subject: Re: [ovirt-devel] Custom Properties code duplication
>>>>
>>>> Hey guys (and gals),
>>>>
>>>> A few patches are up for review starting at:
>>>> http://gerrit.ovirt.org/#/c/27383
>>>>
>>>> In total, about 250 lines of code removed, hopefully not at the cost of
>>>> regression. I put down some reviewers as I saw fit, but everyone can
>>>> feel free to add themselves. Summary of what was done compared to the
>>>> original plan:
>>>>
>>>> 1. Removed said dependencies, except for DeviceCustomPropertiesUtils
>>>> that was using the capture groups features of Pattern, and thus remained
>>>> in the utils project.
>>>>
>>>> 2. Done.
>>>>
>>>> 3. Done.
>>>>
>>>> 4. Almost didn't touch this, it seems to involve a lot of moving parts.
>>>>
>>>> Yours, Lior.
>>>>
>>>> On 30/04/14 18:01, Vojtech Szocs wrote:
>>>>> Hi Lior, please find my comments inline.
>>>>>
>>>>>
>>>>> ----- Original Message -----
>>>>>> From: "Yair Zaslavsky" <yzaslavs at redhat.com>
>>>>>> To: "Lior Vernia" <lvernia at redhat.com>
>>>>>> Cc: devel at ovirt.org
>>>>>> Sent: Wednesday, April 30, 2014 3:28:06 PM
>>>>>> Subject: Re: [ovirt-devel] Custom Properties code duplication
>>>>>>
>>>>>>
>>>>>>
>>>>>> ----- Original Message -----
>>>>>>> From: "Lior Vernia" <lvernia at redhat.com>
>>>>>>> To: devel at ovirt.org
>>>>>>> Sent: Wednesday, April 30, 2014 3:52:16 PM
>>>>>>> Subject: [ovirt-devel] Custom Properties code duplication
>>>>>>>
>>>>>>> Hello,
>>>>>>>
>>>>>>> While adding network custom properties towards oVirt 3.5, I got to take
>>>>>>> a good look at the custom properties code in the backend and frontend.
>>>>>>> It seems to me like there's a lot of code duplication, and I would like
>>>>>>> to suggest the following refactoring:
>>>>>>>
>>>>>>> 1. Remove dependencies on Pattern/Matcher and ApacheCommons methods
>>>>>>> from
>>>>>>> *CustomPropertiesUtils.java, to make them compilable with GWT, and move
>>>>>>> these utilities to the common package. The usage of the said methods is
>>>>>>> minimal and could easily be replaced with String methods, etc.
>>>>>
>>>>> +1
>>>>>
>>>>>>
>>>>>>
>>>>>> In general I am in favor, but how are you going to perform the regex
>>>>>> validation of values?
>>>>>> for example , with vm custom properties, you have - sap_agent that can
>>>>>> be
>>>>>> either true or false.
>>>>>> So you need to validate both at the client and the engine, right?
>>>>>
>>>>> Lior mentioned above the possibility of using String methods, I assume
>>>>> by this he means java.lang.String.matches(String) and similar methods.
>>>>>
>>>>> During GWT compilation, JRE standard String implementation is replaced
>>>>> by emulated (GWT-friendly) String implementation, which implements
>>>>> methods like matches(String) using JavaScript RegExp object. You can
>>>>> find this emulated implementation here:
>>>>>
>>>>>   gwt-user-{version}-sources.jar/com/google/gwt/emul/java/lang/String.java
>>>>>
>>>>> Another alternative is to use GWT's built-in regex support through
>>>>> com.google.gwt.regexp.shared.RegExp class. GWT's RegExp class has two
>>>>> implementations, default one using JRE Pattern/Matcher, emulated one
>>>>> using JavaScript RegExp object. The advantage is (mostly) consistent
>>>>> regex support on both client and server, the disadvantage is server's
>>>>> dependency on GWT JAR. (I don't think we want this.)
>>>>>
>>>>> For simple regex matches, I'd suggest to simply go with String approach.
>>>>>
>>>>> For complex regex matches, we can use JRE Pattern/Matcher on server,
>>>>> and emulate given implementation using GWT RegExp on client.
>>>>>
>>>>> Note that there are (slight) differences between JRE's Pattern/Matcher
>>>>> and JavaScript's RegExp object syntax/behavior. Check GWT RegExp class
>>>>> Javadoc to see details (for simple cases, it's not a big deal, though).
>>>>>
>>>>>>
>>>>>>>
>>>>>>> 2. Modify KeyValueModel to delegate to the common utilities instead of
>>>>>>> duplicating code, e.g. for validation.
>>>>>
>>>>> +1
>>>>>
>>>>> I see that KeyValueModel uses RegexValidation class that delegates to
>>>>> compat's Regex class. Just like above, default Regex class utilizes
>>>>> JRE Pattern/Matcher but client uses emuluated implementation:
>>>>>
>>>>>   gwt-extension/src/main/java/org/ovirt/engine/ui/uioverrides/org/ovirt/engine/core/compat/Regex.java
>>>>>
>>>>> and this emulated implementation uses GWT RegExp class.
>>>>>
>>>>>>>
>>>>>>> 3. Move some validation, which is relevant to all custom properties
>>>>>>> (e.g. duplicate keys), from specific utils (e.g. VmPropertiesUtils) up
>>>>>>> to the shared CustomPropertiesUtils.
>>>>>
>>>>> +1
>>>>>
>>>>>>>
>>>>>>> 4. Optionally change the implementation of custom properties members in
>>>>>>> entities (e.g. VM) from String to Map<String, String>, which would
>>>>>>> obviate the need for different conversion methods between String/Map -
>>>>>>> (de)serialization would only be required in DB interaction.
>>>>>
>>>>> +1
>>>>>
>>>>>>
>>>>>> 3,4 Agreed - good points.
>>>>>>
>>>>>>>
>>>>>>> The main argument against this would be that the frontend is to be
>>>>>>> moved
>>>>>>> over the REST, and might not be written in Java much longer anyway.
>>>>>
>>>>> :) I think there's a misunderstanding regarding our move to REST API.
>>>
>>> I think there isn't any misunderstanding here. I think that common code is
>>> not the best practice here, as Lior mentioned briefly.
>>> IMO one of the main reasons of using the REST is repo/project separation,
>>> some points to consider:
>>>
>>> 1) who will maintain the common code, both UI and core maintainers, we're
>>> missing the point, no?
>>
>> That's not how I see it. Common code is exactly that - common. If the
>> validation is the same in the backend and in the frontend (which it
>> often is), then it's better to share the one library between them than
>> write the code twice.
>>
>> The alternative would be to query the backend for validation, but in
>> some contexts it just wouldn't be responsive enough (e.g. dragging in
>> setup networks).
> 
> As i see it, there are JS (client) validations server validations, they are inherently different.
> For responsive validations you have JS validations (duplicated as you call it)/ or you can query the server for generated JS file (probably not going to happen).
> You didn't answer my question about maintainers.
> 

Who the maintainers are is a minor detail in my opinion, I didn't even
realise this was your main point. They could stay backend maintainers
and approve of fixes that make stuff usable by the frontend.

In my opinion the validations are inherently similar. If we take the
setup networks dialog as an example, the rules for what comprises a
valid configuration are the same, so there's no good reason for all the
logic to be duplicated (even though it is). Same goes for character
validation when entering the name of a new entity - usually the same
regex is checked in the backend and the frontend (although it's copied
and pasted rather than shared).

>>
>>> 2) no dependency separation in UI code.
>>
>> You could still compile the UI without the engine and the engine without
>> the UI and everything would work fine. They just both depend on the same
>> library (but it doesn't have to be the same version in both).
> 
> According to what you're saying, your patches should get rejected, because you move the code to Common project and that's what we're trying to avoid.
> 

Please point me to where I said I thought we should avoid moving code to
common.

>>
>>> 3) complexity.
>>
>> This is part of what having common libraries aims to reduce.
> 
> what is 'common' lib btw?, backend/manager/modules/common is the wrong answer.
> 

To me common is the right answer. It hosts all the common business
entities that are used by the backend, the frontend and the REST. I've
used some networking constants used for backend validation in frontend
validation. And now the custom properties validation is hosted there and
shared.

>>
>>> 4) compatibility.
>>
>> As mentioned above, I don't see how compatibility is compromised.
> 
> As I said, +1 for your patches, because they're already submitted, and they're good for the very immediate future (fyi, regressions is sth that I didn't consider, this will be yours and the reviewers responsibility), but note that the code will be duplicated soon, and altered to meet REST entities; all of this will happen in the near future (near is relative :-)).
> 

That's not what I understood about the move over REST. I thought another
layer of code would be added between the frontend and backend to
translate backend entities to frontend entities if needed, and that all
the current frontend code would stay as is. We might over time pick at
the Java code and incrementally replace it with JavaScript code where
beneficial, but again this doesn't mean duplication.

> Thanks,
> Gilad.
> 
>>
>>>
>>> In conclusion, I don't understand the motivation, but... as I commented in
>>> the patch... +1.
>>>
>>> Thanks,
>>> Gilad.
>>>
>>>>>
>>>>> The plan is to have JavaScript SDK for working with REST API (so that
>>>>> any JavaScript client can work with Engine, be it web application in
>>>>> browser, server application on node.js etc.), and for our GWT-based UI,
>>>>> generate GWT/Java overlay code that delegates (via JSNI) to JavaScript.
>>>>>
>>>>> This is similar approach to projects like SmartGWT which are simply
>>>>> overlays (wrappers) to native JavaScript library (like SmartClient).
>>>>>
>>>>> So the frontend code will still be GWT/Java, it will just consume
>>>>> JavaScript SDK via the above mentioned overlay code for seamless
>>>>> SDK user experience in GWT/Java context.
>>>>>
>>>>> We may think of implementing parts of UI in JavaScript, though.
>>>>> For example, utilizing UI plugins to implement different UI parts
>>>>> as plugins, possibly by using JavaScript directly (which could use
>>>>> JavaScript SDK). However, this is something different and requires
>>>>> deeper thought.
>>>>>
>>>>>>>
>>>>>>> However, to my understanding, there's some time until these changes
>>>>>>> take
>>>>>>> effect. And even if the frontend is to be written in JavaScript, at
>>>>>>> least initially the existing frontend code will have to still be used
>>>>>>> somehow (e.g. auto-translated to JavaScript). That is to say, this
>>>>>>> refactoring might still be beneficial for the not-so-short term.
>>>>>
>>>>> I think this refactoring will be beneficial also in long term :)
>>>>>
>>>>>>
>>>>>> I agree with you here.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>>
>>>>>>> Before going through with this, I wanted to ask for your thoughts and
>>>>>>> to
>>>>>>> hear any specific objections to the proposed changes.
>>>>>>>
>>>>>>> Yours, Lior.
>>>>>>> _______________________________________________
>>>>>>> Devel mailing list
>>>>>>> Devel at ovirt.org
>>>>>>> http://lists.ovirt.org/mailman/listinfo/devel
>>>>>>>
>>>>>> _______________________________________________
>>>>>> Devel mailing list
>>>>>> Devel at ovirt.org
>>>>>> http://lists.ovirt.org/mailman/listinfo/devel
>>>>>>
>>>> _______________________________________________
>>>> Devel mailing list
>>>> Devel at ovirt.org
>>>> http://lists.ovirt.org/mailman/listinfo/devel
>>>>
>>

More information about the Devel mailing list