Create read only copy if needed when opening a store path

[maxrjones]

This PR makes a read-only copy of a store when using StorePath.open(mode="r").

See discussion in #3068. This augments #3068 (it still errors when attempting to open a writeable store path using a read-only store). This is an alternative to #3155.

Closes #3147

TODO:

• Add unit tests and/or doctests in docstrings
• Add docstrings and API docs for any new/modified user-facing classes and functions
• New/modified features documented in docs/user-guide/*.rst
• Changes documented as a new file in changes/
• GitHub Actions have all passed
• Test coverage is 100% (Codecov passes)

[maxrjones]

I moved the mode checking up to the start of the function.

[dstansby]

Looks good to me. Requesting changes because:

• This should get a release note
• My old release note from when I added the error should be deleted

[dstansby]

Suggested change

	ANY_ACCESS_MODE: Final = "r", "r+", "a", "w", "w-"
	ANY_ACCESS_MODE: Final = get_args(AccessModeLiteral)

It might be possible to de-duplicate here (untested) using typing.get_args?

[d-v-b]

at least on my machine, the type inference machinery thinks get_args(...) returns tuple[Any,...]. So this won't work.

[dstansby]

Does it matter what the type is of ANY_ACCESS_MODE though, since it's only used for runtime checking? Not a big deal given these two lines of code are next to each other though, so should be obvious if they end up out of sync.

[d-v-b]

it does matter, because when you annotate a tuple of strings as Final, type checkers will know that it's Literal[*stuff_in_the_tuple], which means you can use that variable for "compile-time" and runtime type checks.

[maxrjones]

It worked on my machine and I was curious what the CI would say so I pushed 7ad760f (#3156) (would be easy to revert again)

[d-v-b]

CI will pass because the return type of get_args() is tuple[Any,...], and so any type is assignable to its contents.

lit_typ = Literal["a", "b"]
lit_val: Final = ("a", "b")

x: lit_typ = lit_val[0]
# the type of x is inferred to be Literal["a"]
y = get_args(lit_typ)[0]
# the type of y is inferred to be Any

[d-v-b]

another argument against get_args here:

y: lit_typ = get_args(lit_val)[15]
# mypy is OK with this
z: lit_typ = lit_val[15]
# mypy sees the indexing error

[maxrjones]

thanks for the explanation, I reverted the change in a1949f1 (#3156).

It seems like our options are redundancy or using get_args inside StorePath.open(). @d-v-b previously suggested that the redundancy option offers minor performance benefits.

[d-v-b]

yeah it's a handful strings, so writing it down twice in essentially the exact same location is not expensive. If we are really worried that the type and the final variable might diverge, we can test for that (using get_args).

[maxrjones]

yeah it's a handful strings, so writing it down twice in essentially the exact same location is not expensive. If we are really worried that the type and the final variable might diverge, we can test for that (using get_args).

this seems like a reasonable compromise. I added a test in 202d606 (#3156).

[maxrjones]

This should get a release note
My old release note from when I added the error should be deleted

Thanks, done in 10e50e4 (#3156)