This change is a preparation for another change that makes all callback
types return a Go error instead of an error code / an integer. That is
going to make make things a lot more idiomatic.
The reason this change is split is threefold:
a) This change is mostly mechanical and should contain no semantic
changes.
b) This change is backwards-compatible (in the Go API compatibility
sense of the word), and thus can be backported to all other releases.
c) It makes the other change a bit smaller and more focused on just one
thing.
Concretely, this change makes all callbacks populate a Go error when
they fail. If the callback is invoked from the same stack as the
function to which it was passed (e.g. for `Tree.Walk`), it will preserve
the error object directly into a struct that also holds the callback
function. Otherwise if the callback is pased to one func and will be
invoked when run from another one (e.g. for `Repository.InitRebase`),
the error string is saved into the libgit2 thread-local storage and then
re-created as a `GitError`.
This change introduces the file deprecated.go, which contains any
constants, functions, and types that are slated to be deprecated in the
next major release.
These symbols are deprecated because they refer to old spellings in
pre-1.0 libgit2. This also makes the build be done with the
`-DDEPRECATE_HARD` flag to avoid regressions.
This, together with
[gorelease](https://godoc.org/golang.org/x/exp/cmd/gorelease)[1] should
make releases safer going forward.
1: More information about how that works at
https://go.googlesource.com/exp/+/refs/heads/master/apidiff/README.md
An Object should be about representing a libgit2 object rather than
showing which methods it should support.
Change any return of Object to *Object and provide methods to convert
between this and the particular type.
The library stores error information in thread-local storage, which
means we need to make sure that the Go runtime doesn't switch OS
threads between the time we call a function and th time we attempt to
retrieve the error information.
Provide a manual way of freeing objects, but set finalizers for them
in case the user does not want to worry about memory management, which
would be useful for commits or trees, which sare typically small.
When the objects are freed manually, the finalizer is unset to avoid
double-freeing, mimicking what the go runtime does.