From 9f9045123efefbd11dd397b4d59596290765feec Mon Sep 17 00:00:00 2001 From: "tompng (tomoya ishida)" Date: Sun, 11 Apr 2021 19:04:31 +0900 Subject: [PATCH] st.c: skip all deleted entries [Bug #17779] Update the start entry skipping all already deleted entries. Fixes performance issue of `Hash#first` in a certain case. --- benchmark/hash_first.yml | 11 +++++++++++ st.c | 9 +++++++-- 2 files changed, 18 insertions(+), 2 deletions(-) create mode 100644 benchmark/hash_first.yml diff --git a/benchmark/hash_first.yml b/benchmark/hash_first.yml new file mode 100644 index 0000000000..c26df1a7ed --- /dev/null +++ b/benchmark/hash_first.yml @@ -0,0 +1,11 @@ +prelude: | + hash1 = 1_000_000.times.to_h { [rand, true]} + hash2 = hash1.dup + hash2.keys[1..100_000].each { hash2.delete _1 } + hash2.delete hash2.first[0] + +benchmark: + hash1: hash1.first + hash2: hash2.first + +loop_count: 100_000 diff --git a/st.c b/st.c index 5b178f6b69..9919f0a734 100644 --- a/st.c +++ b/st.c @@ -1244,8 +1244,13 @@ update_range_for_deleted(st_table *tab, st_index_t n) { /* Do not update entries_bound here. Otherwise, we can fill all bins by deleted entry value before rebuilding the table. */ - if (tab->entries_start == n) - tab->entries_start = n + 1; + if (tab->entries_start == n) { + st_index_t start = n + 1; + st_index_t bound = tab->entries_bound; + st_table_entry *entries = tab->entries; + while (start < bound && DELETED_ENTRY_P(&entries[start])) start++; + tab->entries_start = start; + } } /* Delete entry with KEY from table TAB, set up *VALUE (unless