15:09
<ljharb>
generally, git commits are supposed to be atomic, and often, PRs are also intended to be a single atomic change. If you want to combine multiples in a single PR instead of splitting them into a separate one, then presumably that's because it's more reviewable that way. Review needs to happen in the infinite future as well, and while git commits last forever, github won't.
15:09
<ljharb>
ecma is very concerned with long-term archival, so anything that's not stored in actual git won't necessarily meet that bar (altho they do try to scrape github occasionally)
15:12
<ljharb>
(separately, typically repo maintainers, not PR authors, ultimately decide how to group commits on a PR - 262 is a special case in a number of ways tho, ofc)
15:31
<bakkot>
ljharb re: archival, don't pr refs remain forever?
15:32
<ljharb>
in github
15:32
<bakkot>
in the git repo on github, I mean
15:32
<ljharb>
and in local repos that opt in to downloading them. but you're right that would be a path.
15:33
<jmdyck>
If ecma is very concerned about long-term archival, they should be scraping TC39's repos thoroughly, so that the possibility of github disappearing does not pose a loss-of-archive threat. In which case, TC39 should be able to treat its github repos as archival. And so infinite review of PR-only commits should be possible.
15:40
<jmdyck>
(Also, by "atomic", I assume you don't mean it in the DB sense (git takes care of that), but rather in the sense of leaving the repo (spec) in a valid/consistent/reasonable state.)
17:52
<ljharb>
i mean it in the conceptual sense, yes
17:53
<ljharb>
it's actually a common view that a PR should only be a single conceptual change - i don't subscribe to that philosophy personally, and i'm fine with PRs containing multiple conceptual changes (ie, commits), which is why i don't indiscriminately squash PRs down to one commit (which i think is destructive)
19:09
<jmdyck>
My point is that even when you have a PR that is a single conceptual change, it can still be useful to view it as a set of smaller conceptual changes. I'm disagreeing with the idea that (in any given case) there's only one level-of-detail that's deserving of archiving.