On 4/3/2018 8:00 AM, Derrick Stolee wrote:
There are several commit-graph walks that require loading many commits
but never walk the trees reachable from those commits. However, the
current logic in parse_commit() requires the root tree to be loaded.
This only uses lookup_tree(), but when reading commits from the commit-
graph file, the hashcpy() to load the root tree hash and the time spent
checking the object cache take more time than parsing the rest of the
commit.
In this patch series, all direct references to accessing the 'tree'
member of struct commit are replaced instead by one of the following
methods:
struct tree *get_commit_tree(struct commit *)
struct object_id *get_commit_tree_oid(struct commit *)
This replacement was assisted by a Coccinelle script, but the 'tree'
member is overloaded in other types, so the script gave false-positives
that were removed from the diff.
After all access is restricted to use these methods, we can then
change the postcondition of parse_commit_in_graph() to allow 'tree'
to be NULL. If the tree is accessed later, we can load the tree's
OID from the commit-graph in constant time and perform the lookup_tree().
On the Linux repository, performance tests were run for the following
command:
git log --graph --oneline -1000
Before: 0.83s
After: 0.65s
Rel %: -21.6%
Adding '-- kernel/' to the command requires loading the root tree
for every commit that is walked. There was no measureable performance
change as a result of this patch.
This patch series depends on v7 of ds/commit-graph.
Derrick Stolee (3):
commit: create get_commit_tree() method
treewide: use get_commit_tree() for tree access
commit-graph: lazy-load trees
This patch series is also available as a GitHub pull request [1]
[1] https://github.com/derrickstolee/git/pull/4
