The greppable-invariants check is planned to refuse a missing licence file, and the
release checklist requires one before the first tag. Nothing in this plan creates
it, so two Done-whens currently point at a file that no issue produces.
This issue produces it. It does not choose the licence. That choice is the first
entry on my question issue and stays there, and this issue is unstartable
until it is answered.
What lands once the answer exists.
A LICENSE file at the repository root carrying the full text of the chosen licence,
taken from its canonical source rather than reproduced from memory. A licence trimmed
for length or retyped is a different licence from the one its name claims, and
everybody downstream is entitled to the exact text.
The licence declared in the repository metadata as well, so the sidebar and the API
report what the file says. A root file saying one thing and metadata saying nothing is
how a reader ends up guessing, and guessing about terms is the failure this whole
entry exists to end.
A line in README.md naming the licence and pointing at the file, next to the
intended-use notice that is already linked there.
The second half of entry one answered in the same change. If the runner and the
experiment content end up under different licences, this issue lands both files and
the layout decision says which tree each one governs. If they end up under the same
licence, one file governs everything and the decision record says that was a choice
rather than something nobody looked at.
One limit worth writing at the file rather than only here. Adding a licence does not
place the commits that came before it under those terms. Everything committed until
this lands arrived with no inbound terms at all, and repairing that is a question for
whoever holds the copyright in those commits rather than something a new file in the
root settles. Say so in the decision record, because the alternative is a reader
assuming the whole history is covered.
Done when LICENSE exists at the repository root, its text matches the canonical text
of the licence recorded on the question issue apart from placeholders that licence
itself asks to be filled in, the repository metadata reports that licence, README.md
names it and links it, and the licence leg of the greppable-invariants check passes
instead of printing that it was not asked.
The greppable-invariants check is planned to refuse a missing licence file, and the
release checklist requires one before the first tag. Nothing in this plan creates
it, so two Done-whens currently point at a file that no issue produces.
This issue produces it. It does not choose the licence. That choice is the first
entry on my question issue and stays there, and this issue is unstartable
until it is answered.
What lands once the answer exists.
A
LICENSEfile at the repository root carrying the full text of the chosen licence,taken from its canonical source rather than reproduced from memory. A licence trimmed
for length or retyped is a different licence from the one its name claims, and
everybody downstream is entitled to the exact text.
The licence declared in the repository metadata as well, so the sidebar and the API
report what the file says. A root file saying one thing and metadata saying nothing is
how a reader ends up guessing, and guessing about terms is the failure this whole
entry exists to end.
A line in
README.mdnaming the licence and pointing at the file, next to theintended-use notice that is already linked there.
The second half of entry one answered in the same change. If the runner and the
experiment content end up under different licences, this issue lands both files and
the layout decision says which tree each one governs. If they end up under the same
licence, one file governs everything and the decision record says that was a choice
rather than something nobody looked at.
One limit worth writing at the file rather than only here. Adding a licence does not
place the commits that came before it under those terms. Everything committed until
this lands arrived with no inbound terms at all, and repairing that is a question for
whoever holds the copyright in those commits rather than something a new file in the
root settles. Say so in the decision record, because the alternative is a reader
assuming the whole history is covered.
Done when
LICENSEexists at the repository root, its text matches the canonical textof the licence recorded on the question issue apart from placeholders that licence
itself asks to be filled in, the repository metadata reports that licence,
README.mdnames it and links it, and the licence leg of the greppable-invariants check passes
instead of printing that it was not asked.