Tilde encoding: use ~ instead of - for dash-encoding

[simonw]

Refs #1439

[simonw]

The problem with the dash encoding mechanism is that it turns out dashes are used in a LOT of existing Datasette instances - much of https://fivethirtyeight.datasettes.com/fivethirtyeight for example, and even https://datasette.io/ itself: https://datasette.io/dogsheep-index

It's pretty ugly to force all of those to change to their dash-encoded equivalent - and in fact it broke https://datasette.io/ in a subtle way:

• Search page is broken - 500/404 datasette.io#94

I'm going to try using ~ instead and see if that works as well and causes less breakage to existing sites.

[simonw]

Asked about this on Twitter:

Anyone ever seen a proxy or other URL handling system do anything surprising with the tilde "~" character?

I'm considering it as an escaping character, in place of "-" as described in

Replies so far seem like it should be OK - Apache has supported this for home directories for a couple of decades now without any problems.

[simonw]

Relevant: https://datatracker.ietf.org/doc/html/rfc3986#section-2.1

      reserved    = gen-delims / sub-delims

      gen-delims  = ":" / "/" / "?" / "#" / "[" / "]" / "@"

      sub-delims  = "!" / "$" / "&" / "'" / "(" / ")"
                  / "*" / "+" / "," / ";" / "="

Notably ~ is not in either of those lists.

[simonw]

And in https://datatracker.ietf.org/doc/html/rfc3986#section-2.3 "Unreserved Characters":

  unreserved  = ALPHA / DIGIT / "-" / "." / "_" / "~"

[simonw]

Updated test:

@pytest.mark.parametrize(
    "original,expected",
    (
        ("abc", "abc"),
        ("/foo/bar", "~2Ffoo~2Fbar"),
        ("/-/bar", "~2F-~2Fbar"),
        ("-/db-/table.csv", "-~2Fdb-~2Ftable~2Ecsv"),
        (r"%~-/", "~25~7E-~2F"),
        ("~25~7E~2D~2F", "~7E25~7E7E~7E2D~7E2F"),
    ),
)
def test_tilde_encoding(original, expected):
    actual = utils.tilde_encode(original)
    assert actual == expected
    # And test round-trip
    assert original == utils.tilde_decode(actual)

[simonw]

I've made a real mess of this. I'm going to revert Datasettemain back to the last commit that passed the tests and try this again in a branch.

[simonw]

The state I had got to prior to that revert is in https://github.com/simonw/datasette/tree/issue-1657-wip

[simonw]

The thing that broke everything was this change:

I'm going to bring back the horrible get_format() method for the moment, with its weird mutations of the args object, then try and get rid of it again later.