Re: [GSoC Patch 1/3] commit: introduce helpers for generation slab
From: Derrick Stolee <hidden>
Date: 2020-06-04 14:36:54
On 6/4/2020 3:27 AM, Abhishek Kumar wrote:
quoted hunk ↗ jump to hunk
The struct member generation refers to "generation number" (or more broadly, a reachablity index value) used by commit-graph to reduce time taken to walk commits. However, generation is not useful in other contexts and bloats the struct. Let's move it to a commit-slab and shrink the struct by four bytes. Signed-off-by: Abhishek Kumar <redacted> --- commit-graph.c | 27 +++++++++++++++++++++++++++ commit-graph.h | 5 +++++ commit.h | 3 --- 3 files changed, 32 insertions(+), 3 deletions(-)diff --git a/commit-graph.c b/commit-graph.c index e3420ddcbf..63f419048d 100644 --- a/commit-graph.c +++ b/commit-graph.c@@ -87,6 +87,33 @@ static int commit_pos_cmp(const void *va, const void *vb) commit_pos_at(&commit_pos, b); } +define_commit_slab(generation_slab, uint32_t); +static struct generation_slab generation_slab = COMMIT_SLAB_INIT(1, generation_slab); + +uint32_t generation(const struct commit *c) +{ + uint32_t *gen = generation_slab_peek(&generation_slab, c); + + return gen ? *gen : GENERATION_NUMBER_INFINITY; +}
This is good: if we don't have the value, then use INFINITY. In the header file, perhaps include a warning comment that a caller _must_ first parse the commit or else we have no guarantee that the generation slab is populated. This matches the current expectations before accessing the generation member.
+static void set_generation(const struct commit *c, const uint32_t generation)
+{
+ unsigned int i = generation_slab.slab_count;
+ uint32_t *gen = generation_slab_at(&generation_slab, c);
+
+ /*
+ * commit-slab initializes with zero, overwrite this with
+ * GENERATION_NUMBER_INFINITY
+ */
+ for (; i < generation_slab.slab_count; ++i) {
+ memset(generation_slab.slab[i], GENERATION_NUMBER_INFINITY,
+ generation_slab.slab_size * sizeof(uint32_t));
+ }Here is an example where combining the graph_pos and generation slabs into one would be helpful. The only reason the generation would be INFINITY is if graph_pos is COMMIT_NOT_FROM_GRAPH. If the two values are side-by-side, we could just check graph_pos first and return INFINITY instead of paying this initialization cost as the slab grows. I would also like to avoid initializing the slab if there is no commit-graph present. I wonder if we can populate the slab while parsing the commit-graph and check here if the slab is NULL before doing any other logic? (I'm not sure if this is possible, but it would be nice.)
quoted hunk ↗ jump to hunk
diff --git a/commit-graph.h b/commit-graph.h index 4212766a4f..653bd041ad 100644 --- a/commit-graph.h +++ b/commit-graph.h@@ -8,6 +8,10 @@ #include "object-store.h" #include "oidset.h" +#define GENERATION_NUMBER_INFINITY 0xFFFFFFFF +#define GENERATION_NUMBER_MAX 0x3FFFFFFF +#define GENERATION_NUMBER_ZERO 0 + #define GIT_TEST_COMMIT_GRAPH "GIT_TEST_COMMIT_GRAPH" #define GIT_TEST_COMMIT_GRAPH_DIE_ON_LOAD "GIT_TEST_COMMIT_GRAPH_DIE_ON_LOAD" #define GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS "GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS"@@ -137,4 +141,5 @@ void free_commit_graph(struct commit_graph *); */ void disable_commit_graph(struct repository *r); +uint32_t generation(const struct commit *c); #endifdiff --git a/commit.h b/commit.h index 1b2dea5d85..cc610400d5 100644 --- a/commit.h +++ b/commit.h@@ -11,9 +11,6 @@ #include "commit-slab.h" #define COMMIT_NOT_FROM_GRAPH 0xFFFFFFFF -#define GENERATION_NUMBER_INFINITY 0xFFFFFFFF -#define GENERATION_NUMBER_MAX 0x3FFFFFFF -#define GENERATION_NUMBER_ZERO 0
I appreciate that you are able to relocate these constants to a more appropriate location. Thanks, -Stolee