dirs is maintained now.  The breaking changes from dirs 2.0 to 4.0
described in dirs's readme do not affect users of shellexpand.

Manual merge as gitlab.com is getting mystery errors

this converted all references to actual links for ease-of-use

Fix broken links

Closes #1

See merge request !2

this is because "strip_prefix" uses "starts_with" internally, so it does not need to be used twice

Apply clippy suggestions

See merge request !3

chore(README): fix extra character in example

See merge request !4

As per
  https://github.com/netvl/shellexpand/issues/17
I now have ownership of the crate name.

Fixes #5

Fix panic on un-bracketed variable names starting with unicode

Closes #5

See merge request !5

We're going to want the new import semantics or things will get even
more confusing.

This is tidier IMO and will also be workable when we want to handle
conventionally-valid-UTF8 (ie, for Path).

Do this, without making changes, to help git's rename tracking.

This pushes "ignore non-utf-8" up to the caller.

As a side effect, we fix a bug where non-UTF-8 home directories would
produce mangled pathnames.

Nothing uses these yet

These traits don't have methods yet; we'll introduce them one at a
time.

We also introduce the path module here, so we can introduce the new
pieces side-by-side.

This doesn't actually affect the public docs, happily.

Abolish uses of ordinary "push".

We want to convert parts of the input to str.  If it's not valid
str, it's not a variable name.

format! is only useable with strings.

This works even for &RawOsStr.

We want to be able to compare expected outputs with literals, but the
literals have a type that won't compare nicely in assert_eq.

Likewise variable names.

RawOsStr doesn't have .len().  Our kind-of-string-agile code
needs to use this for slicing.

As per
  !6 (comment 1057662066)

As per
  !6 (comment 1057662070)

RFC - Provide path module

See merge request !6

Sort out cargo features

See merge request !7

Prep for release

See merge request !8