Parse POMs with SAX parser

[alexarchambault]

No description provided.

[alexarchambault]

Benchmark results for the new SAX parser (as of 24955d3)

$ sbt 'benchmark/jmh:run -i 20 -wi 20 -f1 -t1'
[info] Benchmark                              Mode  Cnt    Score    Error  Units
[info] ParseTests.parseApacheParent           avgt   20    0.783 ±  0.011  ms/op
[info] ParseTests.parseSparkParent            avgt   20    2.936 ±  0.037  ms/op
[info] ParseTests.parseSparkParentMavenModel  avgt   20    0.692 ±  0.001  ms/op
[info] ParseTests.parseSparkParentXmlDom      avgt   20    2.288 ±  0.002  ms/op
[info] ParseTests.parseSparkParentXmlSax      avgt   20    0.997 ±  0.001  ms/op
[info] ProcessingTests.sparkSql               avgt   20    0.647 ±  0.001  ms/op
[info] ResolutionDomTests.coursierCli         avgt   20    4.235 ±  0.035  ms/op
[info] ResolutionDomTests.sparkSql            avgt   20  187.636 ±  0.389  ms/op
[info] ResolutionTests.coursierCli            avgt   20    3.856 ±  0.021  ms/op
[info] ResolutionTests.sparkSql               avgt   20  168.218 ±  0.384  ms/op
[success] Total time: 4048 s, completed Jan 23, 2019 2:28:56 PM

parseSparkParentXmlDom and parseSparkParentXmlSax are basically String => Project functions, going from the content of org.apache.spark:spark-parent_2.12:2.4.0 to a coursier.core.Project, either via an AST of the XML, or via a SAX-based parser. The SAX parser divides that time by more than 50%.

ResolutionDomTests.sparkSql and ResolutionTests.sparkSql run whole resolutions of org.apache.spark:spark-sql_2.12:2.4.0, hitting metadata fully in memory (no network or file I/O). The gain is only around ~10 % here. This gain is kind of disappointing compared to the one in the former comparison (which looks at a more atomic operation).