You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Add hit-heavy perf scenarios to compare cache vs baseline
The walk benchmark we already had is cache-miss heavy (one walk per AST,
every node visited once), so the identity cache shows up there as a
small overhead rather than a win. The cache is supposed to pay back in
hit-heavy patterns: re-walks of the same tree, repeated child reads at
the root, and translator-style passes that re-enter visited subtrees.
Adds three modes (--mode=rewalk|reread|subtree) and runs each on both
the PR and the baseline so the comparison is apples-to-apples on the
same runner, same corpus.
| tee "$GITHUB_WORKSPACE/packages/mysql-on-sqlite/baseline-native-subtree.txt"
175
+
134
176
- name: Summarize
135
177
if: always()
136
178
working-directory: packages/mysql-on-sqlite
@@ -148,7 +190,7 @@ jobs:
148
190
echo
149
191
echo '| scenario | result |'
150
192
echo '|---|---|'
151
-
for f in php-parse-only.txt php-walk.txt native-parse-only.txt native-walk.txt baseline-native-parse-only.txt baseline-native-walk.txt; do
193
+
for f in php-parse-only.txt php-walk.txt native-parse-only.txt native-walk.txt baseline-native-parse-only.txt baseline-native-walk.txt native-rewalk.txt baseline-native-rewalk.txt native-reread.txt baseline-native-reread.txt native-subtree.txt baseline-native-subtree.txt; do
0 commit comments