[wildfly-dev] Pretty-printing XML validation errors

[Toby Crawley]

I've done some more work on VDX, and have addressed most of the
technical concerns about it from this thread, namely:

* I18N - VDX is now i18n ready, but currently just has a default
  message set in english.
* Displaying the default message with code - the original message is
  now displayed as part of the output, see
  https://gist.github.com/tobias/ec2846a13b6ff656d8e47cbc85071cb1 for
  an example.
* Handling errors that aren't created by ParseUtils - For any
  XMLStreamExceptions that don't get wrapped by ParseUtils, VDX now
  shows that error the best it can. This still doesn't handle errors
  that aren't XMLStreamExceptions, but is progress. This should also
  handle errors thrown from AttributeDefinitions, since they are
  XMLStreamExceptions.
https://gist.github.com/tobias/551e712892b9326096ecc80bb78bc132
  is an example of handling a non-wrapped exception thrown by the
  reader.
* Size of dependencies - VDX now has no dependencies, and is currently
  61K. Removing the dependencies on Apache XMLSchema and Apache
  Commons Lang reduces the total size by 552K.

Existing issues:

* I18N - though it is set up for i18n, it doesn't yet have any
  translations. How many/what languages do we support for WildFly
  currently? What is the process for getting translations done?
* non-XMLStreamException errors - it's currently possible to trigger
  errors that aren't XMLStreamExceptions when parsing
  standalone.xml. A couple of examples are:

  - adding a duplicate extension under server > extensions causes an
    IllegalStateException that occurs when trying to generate the
    XMLStreamException
    (https://gist.github.com/tobias/59d155afe0c88e268b83cb75734353eb)
  - Creating a file-handler with no attributes under server >
    management > audit-log > handlers results in an error about there
    being no add operation at address [], with no other context
    (https://gist.github.com/tobias/34b7d7791e20148004d883238adf16ac). If
    you add a name attribute, you get a different, more helpful
    management operation error
    (https://gist.github.com/tobias/6d47539aa40de90ec45cfaaec83ea15e),
    but neither error points to the xml configuration as the problem.

  Nothing can really be done with VDX to help with these issues, but I
  would be happy to provide PR's to fix them as I find them if there
  is interest.

The question now is - should I keep working on this? Is there enough
interest in this for me to continue it to the point where it can be
used by wildfly-core?

Is there interest in seeing this applied to xml that is provided as
part of a deployment (web.xml, jboss-web.xml,
jboss-deployment-structure.xml, etc) in addition to (or instead of
standalone.xml)? I haven't yet looked at what that would take.

For reference, the source for VDX is available at
https://github.com/projectodd/vdx, and my changes so far to
wildfly-core are in this commit:
https://github.com/tobias/wildfly-core/commit/2654406bfcfed7cbf255db819f094749487e2d10

- Toby

On Tue, Jul 19, 2016 at 7:28 PM, Toby Crawley <tcrawley at redhat.com> wrote:
> On Tue, Jul 19, 2016 at 4:07 PM, Brian Stansberry
> <brian.stansberry at redhat.com> wrote:
>> Comments in-line, except for something I just thought of.
>>
>> All exception and log messages produced will likely need to follow
>> WildFly’s/EAP's i18n standards. Message prefixed with a code, text produced
>> in a way in an i18n manner with a reasonable way to get localized text into
>> the software.
>
> Ah, good point. It shouldn't be difficult to add localized text to the
> output, the biggest cost there would be the time to translate the
> messages, but I'm not familiar with how i18n is done in our projects.
> For the message code, we could print the original exception message at
> the bottom or top of the validation block. That would then use the
> same code as we provide now, and would provide the same message we use
> now with every error. Would that satisfy the message code requirement?
>
>>>
>>> On Jul 19, 2016, at 2:38 PM, Toby Crawley <tcrawley at redhat.com> wrote:
>>>
>>>> On Tue, Jul 19, 2016 at 3:15 PM, Brian Stansberry
>>>> <brian.stansberry at redhat.com> wrote:
>>>>
>>>> The only big concern I have about this is that we’ll get this behavior for
>>>> some failures but not all. And I don’t want to go down the path of trying to
>>>> force every parser to work in a manner such that we consistently get this.
>>>>
>>>
>>> I haven't looked at it too deeply, but it may be straightforward to
>>> alter staxmapper to allow providing an exception generator that would
>>> allow catching more of the cases that the parsers miss.
>>>
>>
>> I’m not sure how big of a problem staxmapper-thrown exceptions are. (I
>> haven’t really thought.)
>
> That's just the first place I saw errors from outside of ParseUtils,
> but I haven't yet started playing with attribute values.
>
>>
>> What I was thinking more about when I wrote my previous post was parsers not
>> using ParseUtils, or sometimes not using it.
>>
>> Also, a lot of XmlStreamException cases are generated from implementations
>> of org.jboss.as.controller.AttributeDefinition, e.g.
>>
>> https://github.com/wildfly/wildfly-core/blob/master/controller/src/main/java/org/jboss/as/controller/SimpleAttributeDefinition.java#L140
>>
>> Parsers are encouraged to invoke methods on AttributeDefinition to validate
>> attribute values. Perhaps though those are better left alone, as the
>> validators are meant to produce useful exception messages.
>>
>
> If we let these fall back to just pointing out the error location with
> the original message, this may be ok (assuming by "the alidators are
> meant to produce useful exception messages" you mean the messages
> produced by the AttributeDefinitions).
>
> <snip>
>
>>>> A minor concern is how big the added dependencies are. (I don’t know.) We
>>>> want to keep WildFly Core small in footprint.
>>>>
>>>
>>> Right now, the only dependencies vdx (31k) has are commons-lang (which
>>> is already a module in WildFly, but not core-feature-pack),
>>> xmlschema-walker (100k), and xmlschema-core (168k). For the rest of
>>> the work, I don't currently see needing any more dependencies.
>>>
>>
>> Thanks. So about 583K including 284K for commons-lang. The current
>> wildly-core-dist-3.0.0.Alpha3.zip is about 16.9MB, so this is fairly
>> substantial.
>>
>
> I'm only using commons-lang for a Levenshtein distance implementation
> - that could certainly be pulled in, and we could drop the 284k for
> commons-lang. It certainly would be nice to keep core under 17MB
> though.
>
> - Toby