Tamas Cservenak opened MRESOLVER-248 and commented
There is ongoing work to fundamentally change DependencyCollector implementation in resolver.
Resolver was from beginning building dependency graph by doing "depth first" (df) approach, and was downloading POMs sequentially during collection.
With work happing under MRESOLVER-240, MRESOLVER-247 and MRESOLVER-7, collector is altered: is doing now "breadth first" (bf), implements skip-reconcile and will have parallel POM download implemented, and maybe even more (like inter collection cache?).
Still, IMHO letting both (original "df" and new "bf") coexists is technically simple, and would allow us not only simpler comparison of the performance of the two, but there is a possibility that there will be no "one size fit all" solution, so having both knives in drawer is just a win-win. Also, this way we do have a "baseline" to compare with easily: the "df" (original) collector vs "bf" collector.
Ultimately, we MAY prove superiority of one over another, essentially leaving only one collector implementation, as in that case this issue (and the "coexistence indirection") should be just removed, and not enlist it in release notes of upcoming 1.8.0 version.
Remote Links:
1 votes, 2 watchers
Tamas Cservenak opened MRESOLVER-248 and commented
There is ongoing work to fundamentally change
DependencyCollectorimplementation in resolver.Resolver was from beginning building dependency graph by doing "depth first" (df) approach, and was downloading POMs sequentially during collection.
With work happing under MRESOLVER-240, MRESOLVER-247 and MRESOLVER-7, collector is altered: is doing now "breadth first" (bf), implements skip-reconcile and will have parallel POM download implemented, and maybe even more (like inter collection cache?).
Still, IMHO letting both (original "df" and new "bf") coexists is technically simple, and would allow us not only simpler comparison of the performance of the two, but there is a possibility that there will be no "one size fit all" solution, so having both knives in drawer is just a win-win. Also, this way we do have a "baseline" to compare with easily: the "df" (original) collector vs "bf" collector.
Ultimately, we MAY prove superiority of one over another, essentially leaving only one collector implementation, as in that case this issue (and the "coexistence indirection") should be just removed, and not enlist it in release notes of upcoming 1.8.0 version.
Remote Links:
1 votes, 2 watchers