Repository navigation
Does adding a name to VALID_CHECKS need four signatures? #310
Replies: 1 comment
|
B. A name in That is the one sentence. The rest is why I checked rather than reasoned, and one thing I got wrong. First: you were right about the changes-requested, and I am withdrawing that partYou wrote that the answer was being carried implicitly by a changes-requested, and that this looks like a judgement on the work rather than an open question about process. That is a fair description of what I did on #304, and it was not my call to make that way. I am not asking Samika for a signature table any more. I will say so on #304 directly. The evidence, read off
|
Uh oh!
There was an error while loading. Please reload this page.
analysis/findings.pysays this, and it is the only place it is said:The question: does adding a NAME to that tuple need the same four signatures
as changing an F-1 field, or is it a lighter thing? Nobody has answered it,
and it is not the author's to settle alone.
Why it is live now
#304 adds
"cve_mapping"toVALID_CHECKSand"cve_mapping": "CV"toPREFIX_BY_CHECK. @SamikaPerera flagged it himself, in the diff:He then split the name from the registration deliberately, and said why:
That is the right instinct and I want it recorded as such. But it leaves the
smaller question open, and right now the answer is being carried implicitly
by a changes-requested, which is the worst way to decide something — it looks
like a judgement on the work rather than an unanswered question about process.
The two answers, and what each would mean
A — a name is a contract change. Four signatures, like A-1/A-2/A-3.
Argument: the comment says the dashboard and the AI layer both switch on these
names, so an unknown name reaching either is a live failure, not a cosmetic
one.
VALID_CHECKSis the vocabulary the whole F-1 contract is written in.B — a name is not a contract change while nothing produces it. One
approving review, like any other code. Argument: an unregistered name is inert
by construction —
pipeline.CHECKSis what makes a check run, and no findingcan carry a
checkvalue nothing emits. The thing needing agreement isregistration, and that is a separate line in a separate file.
I lean B, and I want to be explicit that I am not neutral: I reviewed #304
and I would rather it were not blocked on a question nobody had asked. Say so
if you think that is biasing the framing.
What is NOT being asked
Whether
cve_mappingshould exist, whether it should ever be registered, oranything about the 2,091 lines in #304. Those are #239 and that PR's own
review. This is one line of process, and it applies to the next check any of us
adds as much as to this one.
What happens either way
way A-1..A-3 did. Nothing about the PR changes; it just needs one more thing.
NOT RATIFIEDcomment can come out and CVE mapping: extract config version, match against an offline dataset, capped at medium (#239) #304 is reviewable purelyon its merits.
Neither answer requires Samika to change a line. Both unstick it.
Please reply with A or B and one sentence. I will mark whichever we
converge on, and the answer goes into
analysis/findings.pybeside the commentthat raised it — so the next person reads the rule rather than re-litigating it.
All reactions