Sharing our fixes upstream, 68 threads and what came back

Between 2026-08-29 and 2026-10-12 I left 71 comments on 68 threads in lukilabs/beautiful-mermaid, each saying that a bug someone had reported or fixed there is fixed in zombie-mermaid. This post counts what went out, what came back, and what did not. The short version: 19 of the 68 threads got a reply or a reaction, all of it friendly, and the other 49 got nothing. One report, upstream #149, is still not fixed.

I have no data on whether any of this led to installs. What I can source is below.

How the numbers were counted

Every number in this post comes from read-only queries of the GitHub API:

Comments sent: all issue comments on lukilabs/beautiful-mermaid (repos/lukilabs/beautiful-mermaid/issues/comments) authored by dfadler, 76 in total. I excluded 5 that sit on pull requests I had opened upstream by mistake (gh defaulted to the parent repo; I closed each with an apology). That leaves 71 comments on 68 distinct threads, 37 issues and 31 pull requests, filed by 49 different people.
Waves: by UTC date of the comment, counting each thread once at its first comment.
Replies: a comment by someone other than me on one of those threads, after my first comment there.
Reactions: the reactions endpoint for each of my comments.

An earlier tally of "about 45 threads" turned out to be the count before the 2026-10-07 wave. 25 + 14 + 6 is 45, and the later waves added 19 and 4 more. Another internal note counted that 10-07 wave as 22 comments. Re-counting from the API gives 19, all posted between 15:41 and 15:43 UTC.

What went out

Wave (UTC)	Threads	What the comment said
2026-08-29	25	"Heads up, this is fixed in our fork", with a one-paragraph description of the fix. All 25 were issues.
2026-09-29	14	Pull requests: the fix is in the fork, usually with fork PR, merge commit, first release and a before and after.
2026-10-06	6	Mostly fixes that had just landed (#156, #157, #158, pie charts on #151), plus #128 and the open question on #149.
2026-10-07	19	A second bulk pass over issues and PRs the first wave had missed, each linking the fork PR that fixed it.
2026-10-12	4	Answers on #51 and #105 (more below), and on #34 and #71, which asked whether C4 diagrams exist anywhere.

Two threads got more than one comment: #64 (wave one, then two on 10-12 answering a new repro, covered in the A* post) and #139 (wave two, then a reply to its author).

The comments link to 43 distinct fork pull requests and issues. The wording changed over time. The first wave was a short heads-up. By the 09-29 wave the comments usually carried the fork PR, the merge commit, the first release containing the fix, and a rendering of the author's diagram. Several of the comments from 10-06 onward were drafted by Claude, and say so in their opening line.

What came back

Nobody was unfriendly. Of the 68 threads:

8 got a reply addressed to us. Another person also added an unrelated repro on #64 on 10-10, which I count separately.
15 comments drew a reaction, 16 reactions in total from 14 people: eight thumbs-up, five hearts, two hoorays and one eyes. None were negative.
19 threads in total got a reply or a reaction. 18 people were involved, since four of them did both.

Three people said they would switch, or already had. I link each rather than quote at length.

@Bougerous on #140 wrote that they would be switching.
@aryasaatvik on #135 wrote "will switch to zombie mermaid" and closed their own pull request that minute. The cherry-picking post tells that story.
@richardtallent on #81 wrote that they had already switched.

Others were warm without committing. @yhatt on #73 said they would keep an eye on the fork. @wsnark, @GauBen and @birenroy said thanks. Birenroy wrote that they had not realised upstream was no longer actively accepting pull requests, which is the most useful thing any of these threads told me: some people do not know. #83 was closed by its reporter two days after my comment. They did not say why, so I do not read anything into it.

Those are statements in public comments. I did not check whether anyone installed anything, and three sentences do not make an adoption story.

What did not come back

49 of 68 threads got no reply and no reaction. That is 72%, and it includes most of the 10-07 wave. A bulk comment on a thread whose author left the project months ago reaches nobody. That was the point of the work, but it should not be counted as a conversation.

#149 is not fixed. The issue asks for rendering in Web Workers and Cloudflare Workers. The only reply there before mine said it was a must for streaming. My comment on 10-06 said the fix was not in the fork either, because the SVG renderer imports elkjs statically and lays out on the main thread, and I asked what the reporter needed. Nobody has answered. Since then 5.0.0 made elkjs an optional peer dependency (#1444, from #1370). That was aimed at the bundle-size obstacle I named, but I have not tested it in a Workers runtime, so the issue is not fixed and I am not claiming it is. The revisit is tracked in #1524.

My own upstream pull request has no comments. #142, a parser fix, is open and nobody has looked at it. I also opened five other pull requests there by mistake in August and October (#145, #146, #147, #148, #153), each meant for this repo. They are closed. Each one sent the upstream maintainers a notification for nothing, and it happened five times.

Credit and authorship

For fixes that began as someone's upstream pull request, the practice is the one in the earlier post: cherry-pick with -x where the code allows it so git keeps the author, keep fork-only edits in separate commits, name the author in the changelog, and tell them.

That did not hold everywhere. Three upstream pull requests were ported without anyone telling the authors: #51 (a CLI, by @vinceyyy), and #105 and #106 (both by @rjulius23). The port of #51 was adapted to the fork's API and build rather than cherry-picked, and the 1.2.0 changelog entry credits it by name and links the upstream PR. The #105/#106 changelog entry links both upstream pull requests. The credit was in the changelog from the start. The notice was not sent until 2026-10-12, when I commented on #51 and #105, weeks after the fact. Neither has replied. The #105 comment covers #106 too, so #106 itself has no comment from me. That is tracked in #1518.

What I would do differently

Tell the author when the port lands, not weeks later. The three late notices were the only real misstep. A line in the porting checklist is enough.
Prefer a thread with a living participant. The replies came from people who were still watching: PR authors, recent reporters, someone who had posted days earlier. A thread whose reporter has not posted in a year is unlikely to answer. This is a hunch from 19 of 68, not a measured rate.
Put the before and after in the comment. The 09-29 wave had a rendering of the author's own diagram and drew a reply or reaction on 4 of its 14 threads. The first wave had a text description and drew a reply or reaction on 7 of 25. That is too small to say the screenshots made the difference, so I am only reporting it.
Say what is not fixed. The #149 comment was the least popular and the most honest. It is also the one I can point back to.
Do not scale this up for its own sake. The 10-07 wave cost 19 comments in two minutes and produced five responses. I would do the same again for threads with a live author, and not otherwise.

The fork-side list is on the fork-fixes page.