ToXmlGenerator ignores Base64Variant while serializing byte[]

[ghenadiibatalski]

Hello, i'm using Jackson mapper and dataformat-xml 2.10.0 brought by Spring Boot 2.2.1.

Currently i recognize following problem: byte[] is properly serialized to base64 but in a wrong Base64Variant. I'm expecting the serialization in a default base64variant MIME-NO-LINEFEEDS but this is not the case, regardless, what variant i configure on an ObjectMapper. This because the ToXmlGenerator accepts the parameter b64variant but doesn't propagate it to underlying _xmlWriter (in my case woodstox stax2) which uses by default MIME Base64 Variant.

Could someone please improve this.

Many thanks in advance and kind regards,
Gena

[cowtowncoder]

Sounds like a reasonable improvement request. I am not sure how easy it would be to do: stax2 API that Woodstox implements does actually have Base64Variant abstraction, as well as pre-defined variants via Base64Variants so it could be straight-forward.

[ghenadiibatalski]

@cowtowncoder thank you for the quick analyzing the problem! I think, it looks pretty straightforward, to map base64 variants from jackson to stax2. The only thing i'm strumbling over are the some convinience methods. So i could break some important use cases while trying to implement a patch without deeper understanding of the library. Also changing the default Variant of stax2 would be a major breaking change, i think (although for my/default use case helpful)

[cowtowncoder]

Yes, there can be challenges. Leaving default choice as-is could be doable (by forcing different default for XmlFactory), and overall change should go in 2.11 probably to reduce risk.

[cowtowncoder]

Starting to work on this, noting that stax2 and jackson default variants align nicely wrt names, but that defaults are different (as already mentioned, so that):

• Jackson defaults to MIME_NO_LINEFEEDS (since lf-escaping in JSON makes this better)
• Stax2 defaults to MIME (since linefeeds not problematic with XML)

so the main challenge is just to figure out a way to retain default setting for backwards compatibility, even if it is not quite optimal for XML use case -- users can change defaults for XmlMapper.
One thing that could be considered for 3.0 would be to change this default; I'll consider that after getting first things done first.

[cowtowncoder]

Oh. Actually looks like Stax2 default of MIME is already used, essentially overriding Jackson-specified default. This would allow keeping the behavior, as long as XmlFactory is configured to force different defaults, which I think should be doable.

[cowtowncoder]

Ok. So, this is now implemented, without changing defaults but making it possible to change default to change serialization.

There are couple of potential concerns that may need future work:

• Custom Base64Variants may be passed but can not really be translated, so it further customizations needed, it may be necessary to add XML-specific overrides so users may pass native Stax2 ByteVariant to use (instead of Jackson's variants that can only map to pre-defined set)
• Reader side does not use Stax2 typed reads (for a few reasons) and hence actually directly uses Jackson's BaseVariant for decoding. If native Stax2 codecs were to be used on writing, reader-side would need some more thought.
• I added tests but noticed that special character encoding of "MODIFIED-FOR-URL" is not checked (I don't expect there to be significant issues there, just observation)

Still, I think this is useful improvement over earlier versions: functionality will be included in 2.12(.0).