In huge repositories, readIdxFile() can be quite slow. Since all index files are already read by OpenRepository(), it's useless to re-parse them when they are needed: we can just re-use what was read the first time.
I noticed this in a program I'm writing, that is essentially iterating over all the objects in the last commit of a specific branch. There are several thousands of objects in several hundreds of subtrees. On my underpowered ARM NAS, this was extremely slow: 2380 seconds on average. go tool pprof showed that about 90% of that time was spent in readIdxFile(). With this patch, the execution only takes 140 seconds on average.
unknwon
commented
10 years ago
In huge repositories,
readIdxFile()can be quite slow. Since all index files are already read byOpenRepository(), it's useless to re-parse them when they are needed: we can just re-use what was read the first time.I noticed this in a program I'm writing, that is essentially iterating over all the objects in the last commit of a specific branch. There are several thousands of objects in several hundreds of subtrees. On my underpowered ARM NAS, this was extremely slow: 2380 seconds on average.
go tool pprofshowed that about 90% of that time was spent inreadIdxFile(). With this patch, the execution only takes 140 seconds on average.