[Tagging] Stop the large feature madness (was: Tag for a plateau or tableland?)
kevin.b.kenny at gmail.com
Fri Apr 19 17:48:51 UTC 2019
On Fri, Apr 19, 2019 at 6:28 AM Christoph Hormann <osm at imagico.de> wrote:
> It is interesting that the idea that large size abstract concepts
> projected onto arbitrarily delineated parts of the physical geography
> by cultural convention like bays, peninsulas, linear rivers and
> plateaus might not be suitable for being recorded in OSM is by several
> people in this discussion reinterpreted as - and i am only slightly
> exaggerating here - that mappers may only record things after they have
> personally touched every centimeter of them.
The fact that as many people 'reinterpreted' your words suggests that
it might behoove you to review them.
In any case, we're in agreement that the Bay of Biscay, the Drake
Passage or the Bight of Benin are quite problematic for arbitrariness.
Something near half of their boundaries are arbitrary - worse in the
case of the Drake Passage. There is some sort of problem involving
these objects that would merit investigation. They are not what I'm
talking about here.
I've at least heard you as arguing that no OSM object ought to have an
arbitrary boundary, anywhere on its perimeter. That rules out almost
all waterways, most of which are connected to other waterways by
rivers or straits. Of course, it rules out peninsulæ and isthmuses.
I know we've had a disagreement before about the Hudson River, where
your interpretation of 'coastline' unambiguously denies the river's
existence for the lower 200+ km of its reach - all because the
boundary between riparian, estuarine and marine environments cannot be
established without drawing an arbitrary line, and the first
unambiguous physical boundary for where it is unquestionably a 'river'
is the Federal Dam in Troy, New York. If locals are consulted, they
will place the river's mouth at the Battery on the southern tip of
Manhattan Island - but somehow, in the quest to destroy all potential
ambiguity, deferring to the locals in this instance is cultural
imperialism. We've been through several proposals for defining the
riparian/estuarine/marine border, for river mouths and complex
coastlines. All have failed on a great many objects in the perception
of the local mappers. Despte your radical intolerance for ambiguity, I
think we do have a broad consensus among the other mappers that the
locals are virtually always right.
I return to the example of the Red Sea. I know of no culture that
denies that it is a single named entity. It has hundreds of km of
coastline, and its borders (except for the mouths of streams, which
are rare in the desert surrounding it) are well defined, with two very
small arbitrary borders: the entrance to the Suëz canal and the narrow
straits on either side of the island of Perim (Ar: Barim: بريم). It
is possible for reasonable people to disagree, perhaps, about whether
the villiage of Faghal (فغال - I may be mistransliterating) is on the
Red Sea or the Persian Gulf, but for hundreds of km, there is no
ambiguity: nobody would disagree that Port Sudan (بور سودان) is a port
on the Red Sea. Mapping the Red Sea as a single point loses that
topologic and shape information. It is throwing the baby away with the
bath water. Whether Faghal is or is not on the Red Sea is a question
that I am happy to leave to any mapper in Faghal who is willing to
offer a local judgment! To the extent to which such judgments trigger
political or cultural conflicts, there will never be a good solution
in OSM. If the locals are currently fighting a war over a boundary,
OSM cannot resolve the question of where the boundary lies.
The Black Sea, of course, offers a similar problem in construction.
Tiny arbitrary lines have to be drawn across the Bosphorus, the Strait
of Kerch, and some river mouths.
I do not accept the idea that the Red Sea is an abstract concept
without a relevant geographic feature bound to it. I do not accept
that "Port Sudan is on the Red Sea, and Odessa is on the Black Sea"
are not verifiable facts. The fact that the seas are joined to narrow
straits across which arbitrary lines must be drawn does not negate the
existence of the Red Sea, or the Black Sea.
It is not just waterbodies and peninsulæ, but even administrative
boundaries have this issue. I live in a state where portions of some
county lines are indefinite - they are in uninhabited wilderness and
have never been surveyed. Nevertheless, the definite portions of the
lines are necessary, as is a consistent topology. For a point in an
inhabited area, where the lines are established, one needs to be able
to compute with certainty, "is this point in Franklin or Essex
County?" Yes, if you choose a point in the uninhabited wilderness, the
answer might be, "I don't know" or might be wrong. But nobody cares
about the answer for an arbitrary point in uninhabited wilderness. A
wrong answer is perfectly acceptable. (It would not be acceptable to
say that the point is in Suffolk County, hundreds of km away, clearly.
But to answer which of two neighbouring counties owns it is likely to
require a series of surveys and a court case.)
We inhabit an ambiguous world. To be intolerant of the slightest
ambiguity is to throw away most of its richness.
I don't support mapping the Red Sea in OSM immediately, because I am
informed that the mapping of objects that large presents severe
technological issues for those that manage OSM's servers and for those
who consume OSM's data at scale. I respect the large-multipolygon
moratorium for that reason. I've also refrained from mapping smaller
objects with ambiguous borders because of the controversy of which
this thread is the latest manifestation. I am expressing my opinion on
the mailing list rather than in the database.
For smaller objects, I'm continuing to muse about introducing linear
ways, tagged something like boundary=indefinite, for completing the
topology of these nearly closed objects. The sole purpose of
boundary=indefinite would be to participate in a multipolygon,
completing its topology, and it would flag to all data consumers that
the precise positioning of that portion of the multipolygon is unknown
or unknowable. I'd further state that a way tagged boundary=indefinite
should have no other tagging, and must always be a way of a
multipolygon. I'm willing to accept a rule of thumb such as 'no more
than 10% of the perimeter of a multipolygon's outer way should be
indefinite' and of course I would state that the judgment of the
locals governs in deciding where to place the endpoints of an
indefinite way. (I insist on being able to bend that rule of thumb to
accommodate very unusual situations such as county lines across
Is this a possible way out of the "all or nothing" attitude that I'm
seeing here? Essentially, your position is, "if we know or can verify
only 99.9% of an area feature's boundary, that's the same as if we can
verify none of it." I'm taking the contrary position that "if the vast
majority of an object's perimeter is known, draw an approximate line
for the rest, mark it as indefinite, and at least benefit from the
part that is known."
More information about the Tagging