GitLab

Names for groups and users must be less than 32. This is a fairly
arbitrary value, just what other OSs use I guess.

Sensitive allocations are now zeroed; any string or buffer that stores a
salt, hash, or password is zeroed before being freed. Added notes in the
docs for password parameters passed in by reference.

Notably also removes the manual implementation of fmt::Debug from user
in favor of deriving it, since auth::Full's implementation of Debug
prevents it from outputting anything sensitive.

More documentation...

Added a GroupBuilder builder pattern for constructing groups instead of
passing the elements as a list of things to AllGroups::add_group. Fields
are verified as allowed names when added to the AllGroups.

All strings are now checked for semicolons before being written to the
database files in `/etc`. Error messages to match. This should prevent
any misuse of the `User`/`Group` fields from yielding a bad system
state.

While putting together a presentation on the way that I set up the
marker types in this crate I came up with a "better" way of doing it,
specifically to save a few words of mem per User by adding the hash to
the type that is given as a parameter instead of as a field of User

Breaks API again, but only for errors (I believe... return types have
become more specific, not less);

0.4

See merge request !35

See the previous commit for caveats on this release. More info in the
tag.

Several peripheral API changes here ramping up to the 0.4.0 release.
- The `All::remove_by_*` functions now return booleans indicating if the
  gruser was removed or not.
- User and Group no longer implement display for use in passwd/shadow
  file serialization; the implementations have moved to private
  functions.
- Parse errors now return a line number (only the line number! Don't
  leak the file contents).
- Invalid lines are no longer ignored during passwd/group file parsing.
  This has potential to cause breakage but I'd estimate that to be a
  pretty small risk. It does now report any parse errors.

Essentially just make the tests pass. Enabling locks on linux causes
deadlocks in the tests (even when running single-threaded?), so I need
to look at that some more.

This also includes a couple of documentation combing things, there will
be another commit to follow this up.

This compiles, but I think the tests are deadlocking somehow, even with
an unlock being called explicitly in the AllGroups drop. More work is
needed to make this work.

Uhg.

- `FromStr` is no longer used for /etc/group parsing
- All the structs in this crate will debug print correctly
- Trailing whitespace/empty user lists for /etc/group entries are now
  parsed correctly. Tests added for those cases as well.

I never liked how I had to pass around a bool and ensure that the user
had used the API correctly and panic if they hadn't. This commit adds a
marker type to `User` and `AllUsers` that ensures that a user of the
library cannot use authentication and other password related methods
unless they've instantiated the user with authentication information.

Got rid of the tests that ensured panics for incorrect usage of the API;
incorrect API use is a compile-time error now.

I removed `FromStr` for `User` and added it as a private method, later
commits will do the same for `Group` and possibly transition away from
`Display`.

Documentation in the works.

Resolves #26.

Since dirs is a reverse dependency of this crate and is used an awful
lot, it's nice to have as small a dependency footprint as possible. dirs
doesn't need the authentication stuff so we can drop a large number of
deps. I imagine this will probably be useful in other contexts as well.

Bump rust-argon2 dependency

Closes #30

See merge request !34

Bump rust-argon2 dependency to 0.7

See merge request !33

Switch from rand_os crate to getrandom

See merge request !32

This extracts part of !30 into it's own PR. getrandom has a smaller dependency footprint, and the version of rand_os being used was quite out of date.

v3.0

See merge request !26

I think we're ready to merge, I'm happy with the state of the code and
the state of the docs. I'd like to majorly refactor before a 1.0.0
release, but this is just fine for now.

Replace rand 0.5 with rand_os 0.1

See merge request !27