mORMot2 Daily - the commit log

flydev · 2026-08-28 18:00:31

Hi, friday afternoon ai driven fun: a small static site that reads the day's commits from synopse/mormot2 upstream and rewrites them for people who don't have the internals in their head - what changed, what it means for an existing Delphi/FPC backend, and whether you actually need to do anything about it.

https://github.com/flydev-fr/mormot2-daily-dev or https://mormot2daily.fsb.dev/

How it works: every morning a job pulls the new commits with their diffs, an AI agent reads them and writes a structured summary, and that summary is schema-validated before anything gets published. There's an RSS feed if you'd rather read it with your coffee and a JSON feed.

The caveat: the summaries are machine-written and can be wrong. Every entry links to its commit, and the commit is the source of truth. Entries are tagged *interpretation* when the reviewer is reasoning about consequences rather than reporting what the diff actually says, and *auto-classified* when a day didn't get reviewed at all. You should know which sentences to trust.

Today's edition, for a taste: the async thread wake-up went lockless, `TOrmTable.GetAsVariant` quietly changed from `out` to `var`, and `Lecuyer` moved units.

If it's useful I'll keep it running. It will be improved from time to time.

Have a nice weekend

Last edited by flydev (2026-08-29 06:59:54)

ab · 2026-08-28 22:14:20

I like it very much.

I will put a link to my boss so that he knows what I did everyday.

The first link seems down.
The second is great. Just one mistake: the "var variant" parameters are cleaned by the function itself so change at all for the caller: just slightly faster.

zen010101 · 2026-08-29 03:52:17

nice work! I like it very much too

flydev · 2026-08-29 07:06:13

I fixed the gitHub URL.

I'll add the Chinese version @zen

I will put a link to my boss so that he knows what I did everyday.

flydev · 2026-08-29 08:40:20

I had scheduled another LLM on the same commits, and this morning I got this one: https://github.com/flydev-fr/mormot2-da … 5285473602

imo it's more relevant, especially on the mistake you pointed out - it got the `var variant` part right:
(the #3 - MEDIUM: ORM Variant lifetime leak fixed, plus TOrmTable Variant-generation refactoring)

And it links to your blog post from yesterday. I'll let both run in parallel for a few days and compare before picking one but I think this last is far more specific.

Last edited by flydev (2026-08-29 08:41:45)

zen010101 · 2026-08-29 08:48:36

flydev wrote:

I'll add the Chinese version @zen

That’s awesome, thanks!

ab · 2026-08-29 15:03:07

What about not only a sumup but a code review, to identify any security issue or potential regression?
Perhaps too many false positives...

flydev · 2026-08-30 14:22:32

site update:

the digest is now in 4 languages - english, french, chinese and russian. @zen010101 chinese is in
new UI, denser, and readable on a phone now

what do you think about it?

@ab your correction is in. The model reasoned fine about pascal semantics and just never opened the body of the function.

ab wrote:

What about not only a sumup but a code review, to identify any security issue or potential regression?
Perhaps too many false positives...

I have that in mind since some time but.. two things were a blocker.

first, the false positives, imo gemini would produce far too many on this codebase - the ones that hold up on mormot2 at this level are opus-5, gpt-5.6 and qwen-3.8.
second, no api key on my side. So I will run it on a personal subscription, with opus, through claude-code-action. No idea what that eats in tokens, honestly - that is a good part of what I want to find out.

I also added a custom build of tokensave to the CI, patched so it indexes the framework .inc files and not only the .pas ones. mormot2 keeps ~61k lines in 15 .inc files - the whole posix/windows layer - and without the patch the code graph is blind exactly where it matters. I checked the parse before wiring it (fresh rebase on this morning): mormot.core.os.posix.inc gives 764 routines, os.windows.inc 612, as clean as a normal unit. Should help a lot on "who calls this", which is the question the reviewer keeps needing.

and what filters is not the model, it is the bar:

a finding must quote the code that proves it, verbatim. A script re-reads that file at that sha and rejects the finding when the quote is not in it
a finding must name a public entry point that reaches the defect, with the call quoted too. No path, it goes to an "unverified" bucket instead
then it spends a pass trying to refute its own finding, and writes down what it checked

the workflow (pipeline):

one session per commit, as a matrix - 25 commits, 25 isolated sessions, each with mORMot2 checked out at its own sha, so it can open a whole function body and git grep the callers. Each run returns its own usage, so it gets logged and I can show it.

one more thing on the "potential regression" part. An edition is a window closed at midnight, but the code settles over several days - a change can land one morning and get its own fix a few commits later, which happened this week on the wake-up path rewritten at 16617/16618 and fixed at 16623. A one-day view cannot see that, so recommending the first revision is simply wrong.
The payload now looks ahead: for each commit, the later commits touching the same routines, flagged when they look like a fix - thanks @landrix

so tomorow the next CI job will ran on claude and then next is running some days or a week and posting what it found and what it cost. Misses included.

zen010101 · 2026-08-30 15:42:58

Thanks @flydev — I’ve taken a look at the Chinese version. I really appreciate you adding it and getting it online so quickly!

From my reading experience on a PC, one small detail I noticed was the Chinese font fallback. The bundled Latin fonts do not contain CJK glyphs, so the browser otherwise chooses its own fallback
font.

I prepared a small adjustment using commonly available Chinese system fonts. On my PC, this makes the Chinese text feel more natural and easier to read, while preserving the existing Lekton/
Serrif design and adding no extra font downloads.

I submitted it as a PR here:
https://github.com/flydev-fr/mormot2-daily-dev/pull/9

Thanks again for making the Chinese version available!

flydev · 2026-08-30 16:11:00

Sorry, merged, thanks

sh17 · 2026-09-03 08:01:07

This will be my new browser start page. :-)

flydev · 2026-09-03 08:51:35

yes, this tool is useful!

If any of you have ideas on how to set up the workflows for code review, I'll take them - my attempts fail so far, and it's a cost and logic problem.

"edition": "2026-09-03",
"totals": {
  "commits": 22,
  "sessions": 61,
  "cost_usd": 54.5882,
  "turns": 1198,
  "seconds": 11152
}

$54 and 1198 turns on claude-opus_&_sonnet-5, for one day of commits. 3 hours of model time I should probably go read how Mozilla runs their security tooling minus Mythos llm

PS: it runs on a personal subscription so those dollars are an API-list equivalent, not a bill. Just a way to compare runs.

ab · 2026-09-03 15:27:23

I worked more than 3 hours... :-(

ab · 2026-10-13 09:22:01

My most beloved morning news website https://mormot2daily.fsb.dev/ sounds stuck to last Friday.

Please let the marmot/dinosaur come back to live!

flydev · 2026-10-13 18:10:51

the marmot is alive - it's my automation, not the data.

github has been firing the scheduled workflow 6-8h late since oct 10 to actual ~10-11 UTC. My review step runs at 06:00 and gives up at 06:49, so the payload lands after its own window and then gets skipped the next morning. I wasn't aware that github cron is best-effort.

Fixed it by letting the llm check more often instead of relying only on the cron. It's running and should be back later this evening

ps: And I will be back about my last PR you fixed with some others findings

ab · 2026-10-13 21:34:07

I guess there is something wrong with the commits reading: on https://mormot2daily.fsb.dev/edition/2026-10-12.html i.e. monday it only see 11 commits whereas there was much more than this!

flydev · 2026-10-14 08:32:30

You are right! the edition dates currently represent rolling fetch windows rather than calendar days. Fixing it, so each edition covers one complete calendar day.

mORMot Open Source

#1 2026-08-28 18:00:31

mORMot2 Daily - the commit log

#2 2026-08-28 22:14:20

Re: mORMot2 Daily - the commit log

#3 2026-08-29 03:52:17

Re: mORMot2 Daily - the commit log

#4 2026-08-29 07:06:13

Re: mORMot2 Daily - the commit log

#5 2026-08-29 08:40:20

Re: mORMot2 Daily - the commit log

#6 2026-08-29 08:48:36

Re: mORMot2 Daily - the commit log

#7 2026-08-29 15:03:07

Re: mORMot2 Daily - the commit log

#8 2026-08-30 14:22:32

Re: mORMot2 Daily - the commit log

#9 2026-08-30 15:42:58

Re: mORMot2 Daily - the commit log

#10 2026-08-30 16:11:00

Re: mORMot2 Daily - the commit log

#11 2026-09-03 08:01:07

Re: mORMot2 Daily - the commit log

#12 2026-09-03 08:51:35

Re: mORMot2 Daily - the commit log

#13 2026-09-03 15:27:23

Re: mORMot2 Daily - the commit log

#14 2026-10-13 09:22:01

Re: mORMot2 Daily - the commit log

#15 2026-10-13 18:10:51

Re: mORMot2 Daily - the commit log

#16 2026-10-13 21:34:07

Re: mORMot2 Daily - the commit log

#17 2026-10-14 08:32:30

Re: mORMot2 Daily - the commit log

Board footer