• String table holds every literal value used in the XML.
• Simple BER-like encoding carries the actual tag data (tag name, attributes, length of binary blob.)

  
  I am a salad.

String table:
0. foo
1. bar
2. baz
3. blorp
4. I am a salad.

Structure:
tag (bytes xxx) (name 0)
  attribute (name 1) (value 2)

  tag (bytes 0) (name 3)
  content (value 4)

Reporting the byte size of a container does mean having to predict what will be there or having to buffer up output. Some small clients do not like doing this.

A similar idea could be used to store "compacted RDF" turtles.

Considerations

• FSST is a simple compressor for text strings which turns 1-8 byte strings in to offsets. That could be used if compressing the string table is desired.
• Though just slapping a compressor on a single payload is probably fine too.
• CRIME attacks happen when the whole stream is using the same dictionary; it should be fine if every payload is independently compressed.

• FSST paper