* han/rm-blob-content:
change BlobContent to BlobRef that borrows its content from git2::Blob
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
Rust 1.67 was announced[0].
Update the nix sources to be on par with the latest rustc version.
This required fixes to clippy warnings for using variable
interpolation for format strings.
There was also an advisory error related to libgit2-sys[1], which is
transitively depended on via git2. The fix is included in the 0.16.1
version of git2.
[0]: https://blog.rust-lang.org/2023/01/26/Rust-1.67.0.html
[1]: https://rustsec.org/advisories/RUSTSEC-2023-0003
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
The motivation is to avoid copying the content bytes that could be
potentially quite large.
Also support Blob<BlobRef>::to_owned() just in case.
Signed-off-by: Han Xu <keepsimple@gmail.com>
Moved `Error` in `repo.rs` out into a crate level `Error` in `error.rs`.
Repository specific errors are wrapped by the `crate::Error` as well.
Signed-off-by: Han Xu <keepsimple@gmail.com>
Users would call Repository::last_commit().
Also renamed "id" to "tree_id" in Directory and "blob_id" in File.
Signed-off-by: Han Xu <keepsimple@gmail.com>
The Error type is no longer used thanks to the clean up of the Tree
and Blob constructions now being exposed through Repository.
Remove the Error type and clean up the imports as a result.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
Comparing tree entries would only take into account their names. While
this is useful, it is also useful to compare by their entry kind
first, where trees take precedence over blobs.
A test is added to ensure the ordering is respected. While testing, it
was noticed that the entries aren't sorted when be passed in. When a
`Tree` is constructed it will now sort the entries first, ensure they
are given back in sorted order.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
- Merge `person.rs` into `object/commit.rs`.
- Move `source/commit.rs` into `object/`.
- Merge `file_system/directory.rs` into `file_system.rs`.
- Rename `file_system.rs` to `fs.rs`.
Signed-off-by: Han Xu <keepsimple@gmail.com>
The `stats` method always provided the stats at the `head` of the
repository.
Introduce the `stats_from` method which takes a revision to start
from. `stats` can be defined using `stats_from` and `head`.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
To help frontends with figuring out what kind of line is being talked
about for added and deleted files, the serialization of `Addition` and
`Deletion` always include a type. This mirrors the serialization of
the `Modification` type, which is tagged because it's an enum.
The serialization is hand-rolled since it is desirable to make sure
that added files and deleted files are only ever composed of lines
that are not modified.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
If a git Tree is a subtree then it no longer contains the path
information while traversing it. The traversal will only carry the
path with it if it does not skip directories, which in the case of
`entries`, it does.
To remedy this, the `path` method is used to pass the information to
`from_entry` instead.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
The http-api code expects the 'path' serialized field to be the full
path for trees, tree entries, and blobs, i.e. the path where it is
located concatenated with the name of the entry.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
A placeholder `.expect("TODO")` was left in the diff code while
refactoring.
Change the `expect` to `?` and use the correct `type Error`.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
To have a clear separation of which are public functions and which are
private functions, two `impl Repository` blocks are created.
This lead to some refactoring of names and inclusion of helper
functions, removing the need for the `git2_repo` function.
The `get_commit_stats` function seemed off, since the `Stats` in this
case are to do with the repository. Thus, it does not take a commit
anymore and instead uses the `head` commit to calculate the
repository's `Stats`.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
This is an overhaul of the diff module. The major changes are as
follows.
It shouldn't be possible for an added or deleted file to have any
other kind line changes other than additions or deletions. So there
are now 3 different kinds of changes, Addition, Deletion, or
Modification.
Hunk and Hunks now take a generic type parameter to allow for the
three types of changes. As a consequence a FileDiff does the same.
For the actual file representations, they fill in their parameters as
Addition, Deletion, or Modification.
During this refactor some renaming was also made:
* CreateFile -> Added -- as well as the created field being renamed
to added.
* DeleteFile -> Deleted
* MoveFile -> Moved
* CopyFile -> Copied
* ModifiedFile -> Modified
The Stats are also now carried in the Diff instead of being
calculated. They can be retrieved from the `git2::Diff`.
The final change was to remove the `source::Commit::stats` field in
favour of accessing the `Diff::stats` instead.
The above changes mean that the serialization of `Diff` and `Commit`
are not backwards compatible.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
The View type is specific to Heartwood and was preemptively defined
when in fact it will not be used. YAGNI.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
Remove the syntax feature flag and any of the syntax highlighting
logic. This is not required any more from the clients implementation
and so we minimise this codebase in favour of any downstream dependent
to define how they highlight the blob's content.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
Instead of creating a named feature flag, opt for using the `serde`
package name as the feature flag name. This makes it more clear that
the feature is only enabling serde serialization and nothing else.
This patch also cleans up the testing flags by removing the serialize
flag for testing, as this should just always be tested, and also adds
the syntax flag for future testing.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
Refactor the source data types to reuse existing logic and update to
the latest protocol.
The `Blob` and `Tree` data types are updated to be backed by `File`
and `Directory` respectively. This allows the construction of them to
be much simpler and reuse any existing logic on those types. This also
lead to the removal of `ObjectType` and `Info`.
The `Revision` type was removed in favour of using the `Revision`
trait.
The `Revisions` type was renamed to `View`, while also renaming the
`revision` module to `view`. A `View` is a simple encoding of what a
peer's "view" would look like, i.e. what references do they have. This
also required the introduction of getting free-form categories in the
reference db, i.e. `refs/<category>`, and so a `Categories` iterator
was also introduced.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
This is a straightforward move of all the code source browsing modules
into a module called `source`.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
This patch changes the glob module in a couple of ways, outlined
below.
The first change is the use of PatternString in the constructor
functions. This removes the fallibility of the constructors. This
meant that the tests changed to use the `pattern!` macro instead.
The second change is splitting out the `Glob<Branch>` methods into
`Glob<Local>` and `Glob<Remote>`. This means that the local and remote
globs can be constructed separately, and are then combined via the
`branches` and `and` methods.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
The file_system types, Label & Path, are overengineered. It's more
natural to use the Path and PathBuf types provided by the std library.
This replaces any mention of Label with String and any use of the old
Path with either std::path::Path or std::path::PathBuf, depending on
what kind of ownwership is needed.
Anywhere a "root" directory is needed `PathBuf::new()` is used,
i.e. an empty `PathBuf`. Otherwise `Path::new` can used to pass
directory, or in fact a &str can also be used since it implements
AsRef<Path>.
The `File` and `Directory` types also contain there relative paths,
which can be retrieved while traversing the Tree.
Note the special file name `"special/faux\\path"` requires the use of an
escaping string construction, i.e. `r"special/faux\\path"`.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
This patch applies some refactoring to the directory module. It
applies some renaming and adds some useful methods to the `File` and
`Directory` types.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
The `tag_names` and `branch_names` functions use the `Tags` and
`Branches` functions which require to go through their `TryFrom`
instances. This can be expensive, in the case of `Tags` especially.
Add `TagNames` and `BranchNames` iterators that only return the names
of the tags and branches, respectively.
This requires to change the definition of Revisions, where the
NonEmpty branch list is now foregone and there references are
collected in a BTreeSet to ensure ordering.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
Add `short_name` methods to `Branch`, `Local`, `Remote`, and `Tag` to
get the short version of the reference name.
Also adds a `remote` method to `Remote`.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
A `Namespace` is a '/' separated list of components, where each
component represents another level of the namespaces hierarchy. Notice
that a `RefString` is also a '/' separated list of components, thus it
could be reused to define a `Namespace`.
This change uses this observation and reworks `Namespace` to be backed
by a `RefString`.
This naturally leads to defining its helper functions to return
`Namespaced` and `NamespacedPattern` reference names. They are
reworked to take a `Qualified` for regular reference names and
`QualifiedPattern` for refspec names, respectively.
This leads to changing the backing vector of `Glob` to use
`QualifiedPattern` instead of `PatternString`.
While changing `Glob`, it was also noticed that the PhantomData was
not used correctly -- all `impl`s used a generic parameter. This was
changed to use the `Namespace`, `Tag`, and `Branch` types directly,
providing the type-safety that would be expected.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
The current `error::Error` type aggregated a lot of different error
scenarios that were not for the actual repo actions, instead they
would be for the domains that the repo actions would return.
This is a reorganisation of those by localising error variants in
their modules:
* namespace::Error
* commit::Error
* glob::Error
* adding `type Error` to ToCommit
* adding `type Error` to Revision
The main Error type now consists of these other Error types and lives
in the `repo` module. It is re-exported at the top-level since
RepositoryRef is the main interaction with the library.
During this refactor, some reorganisation of functions was done as
well:
* Move `get_file` from `Commit` to `RepositoryRef` as
`get_commit_file`.
* Have a blanket implementation of `ToCommit` for any `Revision`,
instead of repeating the implementations.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
The aim of this commit is to better represent the reference names
using git-ref-format.
The first places to change were the Branch and Tag types. For Branch,
the previous types were simply wrappers around the names. They stay
this way, using RefString instead. For Tag, the `name` field is
replaced with RefString.
Note: The parsing logic becomes easier here by utilising the
Namespaced and Qualified types.
The consequences of these changes meant that areas that used
BranchName or TagName are changed to use RefString.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
Other VCS systems are no longer expected to be supported.
Remove the git hierarchy up to the top-level and remove the vcs
hierarchy.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
This pragma gets in the way more than it solves anything.
Signed-off-by: Fintan Halpenny <fintan.halpenny@gmail.com>
X-Clacks-Overhead: GNU Terry Pratchett
- Added Oid into Directory and File.
- Moved file content and directory content out into their own.
- Lazily load directory content on-demand.
- Take advantage of git2::tree to retrieve an entry on-demand.
Also added a new example for browsing.
Signed-off-by: Han Xu <keepsimple@gmail.com>
Fintan Halpenny