You do need special considerations for XML files though - there are several solutions
The weakest solution is to rely on the ability of the target user to spot diffs and correctly merge XML files. And also not to use automatic merging, ever, because the nature of XML files means that conflicting changes may not occur in adjacent lines.
The next (and inadequate) solution is to order the XML consistently - you can do this in your diff tool, or you can write your tools to produce a reliably ordered file in the first place.
Many tools that work on XML files exhibit what I call "juggling" - the elements and attributes change order when you change the value of them or their siblings, because the software is directly using the DOM to manipulate the file - and does this by creating new objects and removing the old ones from the collection. This is a real PITA for text-based diff tools because not all the changes will even conflict with each other (element sequences are often spread across multiple lines, more so if you put attributes on their own line to enhance the ability to merge).
So, you can either write your code to write a consistent order - usually by serializing a fresh XML stream from a model when you write the file.
Or you can add a layer that re-orders the document when you diff it - many of the available diff tools will let you do this. For some files, I used to write an XSLT sheet (to re-order elements consistently). For attributes, I wrote an extra option for Tidy that sorts attributes - doing that plus laying them out on separate lines is sufficient for many files. I've gone as far as writing custom tools that unpack HTML written into an attribute (with all the escape sequences that entails) into a CDATA section for clarity, runs it through Tidy, and then repacks everything after you're done.
Intermediate : I've thought of taking this a step further and converting the XML to a directory tree of text files designed to merge well, principally to make things clearer for end-users who currently have the kind of diff-tool-plus-converter described above but still occasionally make merge errors.
The next step is to write tools to specifically diff your model. This is probably a bridge too far for most developers, because we have the kind of brain that can abstract a text representation of the model and map it to the actual model that will be created. For end users, it may well be advisable.
Diff / merge tools are a field that need more work - currently the main users are developers who can cope with them being a bit immature. But we will increasingly see collaborative tools based on the kinds of version control that we take for granted, and normal users will need to be able to do this stuff too.