DOM, SAX or StAX?

DOM, SAX or StAX?

Choosing Between DOM, SAX and StAX

On the 100,000-book sample file of lxml Performance (sold-out books relabeled so that StAX reads to the end, and precompiled so that compilation is not timed):

DOM, SAX and StAX on a 58 MB file, with time and peak memory

~/de-venv/bin/python ../s2-26/make_big.py 100000
sed -i 's/out-of-stock/in-stock/' big-catalog.xml        # so StAX reads to the end
javac -d classes Dom.java Sax.java Stax.java
for api in Dom Sax Stax; do
  /usr/bin/time -f "  %e s, peak RSS %M KiB" java -Xmx2g -cp classes $api big-catalog.xml
done

Output

DOM:  total 2245674.08, b3 now in-stock
  7.93 s, peak RSS 889060 KiB
SAX:  total 2245674.08
  2.89 s, peak RSS 70908 KiB
StAX: total 2245674.08
  3.46 s, peak RSS 95320 KiB

Choosing a parsing model
Need	DOM	SAX	StAX
Memory	Whole tree (15x the file)	Constant	Constant
Random access, edits	Yes	No	No
Stop early	After the full parse	Throw an exception	break

Streaming memory here is mostly the JVM itself; timings come from a shared 4-CPU machine, so compare ratios. A common hybrid streams records with StAX or iterparse and builds a small tree per record.

MENU