From a2717f4a5576f389f92028b6180bbb87b8c414f0 Mon Sep 17 00:00:00 2001 From: Sebastian Kuzminsky Date: Tue, 22 Jul 2025 13:44:05 -0600 Subject: [PATCH] add comment 6ce4f668904c7016c846bcc9c0429aca on issue 87fa3146b90db61c4ea0de182798a0e5 --- .../6ce4f668904c7016c846bcc9c0429aca/description | 15 +++++++++++++++ 1 file changed, 15 insertions(+) create mode 100644 87fa3146b90db61c4ea0de182798a0e5/comments/6ce4f668904c7016c846bcc9c0429aca/description diff --git a/87fa3146b90db61c4ea0de182798a0e5/comments/6ce4f668904c7016c846bcc9c0429aca/description b/87fa3146b90db61c4ea0de182798a0e5/comments/6ce4f668904c7016c846bcc9c0429aca/description new file mode 100644 index 0000000..339771b --- /dev/null +++ b/87fa3146b90db61c4ea0de182798a0e5/comments/6ce4f668904c7016c846bcc9c0429aca/description @@ -0,0 +1,15 @@ +`perf` shows that the majority of the time it takes to run "ent anything" +is spent reading the issue database, and the majority of the time it +takes to read the database is spend running `git log` to find the author +and ctime of issues and comments. + +I've reducted this a bit (in the `speed-test` git branch) by running +`git log` half as many times, but the shape of the flame graph remains +the same. + +We've discussed lazy-loading things to avoid doing this work when we don't +need to. For example, if we're trying to look up the tags off issue A, +there's no need to ingest all the comments of issue B. + +Another option is to store Author and Creation-time the normal way, +as files in the issue directory. We'd never have to run `git log` at all.