fastannotate: add a missing b prefix Spotted while debugging a test failure, but this wasn't the problem. # skip-blame b prefix Differential Revision: https://phab.mercurial-scm.org/D5499

rust-cpython: using the new bindings from Python The Python callers detect if we have cpython or direct-ffi bindings and fallback to the Python implementation if none is present. This intermediate state allows to compare the three possibilities. Differential Revision: https://phab.mercurial-scm.org/D5442

rust-cpython: binding for LazyAncestors The `mercurial.rustext.ancestor` module will not in the foreseeable future be a drop-in replacement for the pure `mercurial.ancestor`, because the Rust variants take the index at instantiation whereas the Python ones take a parents function. From the Python side, using the index from `ancestor` would leak internal details out of `mercurial.revlog`, and that's unwanted. Therefore, given that classes defined in `rust-cpython` have the same names in both language, we keep the Rust naming convention (CamelCase). Eventually, though, the ancestor module can be placed under control of `mercurial.policy`, but it will still be up to `revlog` to be aware of that and play the role of a factory for instantiation. Differential Revision: https://phab.mercurial-scm.org/D5441

discovery: move common heads computation inside partialdiscovery object This remove one of the private attribute access. In additions, head tracking and computation is a typical operation we can speed up using Rust.

discovery: introduce a partialdiscovery object This object will ultimately gather the data about common, undecided and missing revs in a single place and deal with most graph related computations. The goal is both to clarify the algorithm and to help provides a simple and clear API that can be reimplemented in Rust. For now, we only moved the `common` set in the object. In this commit, some direct access to the "private" `disco._common` attribute persist. They have not been removed yet because we won't need to expose a full API identical to `incrementalmissingancestors` and it seems simpler to access the attribute directly until the replacement is in place.

discovery: move handling of sampling special case inside sampling function The handling of cases where the number of revisions to sample is smaller than the sample size can be moved with the sample function themselves. This simplifies main logic, preparing a coming refactoring.

discovery: minor fix to some conditionals Since `size` is the upper limit of the sample, we should include it in the check. Otherwize the `more` variable will be zero and the sampling will be useless

delta: reuse _findsnapshot call from previous stage Two different stage of the sparse-revlog logic needs the _findsnapshot data. To avoid recomputing it twice, make it possible to reuse the first computation in the second step. example affected manifest write before: 0.067141s after: 0.064252s (-5%) (total gain since start of series: 95%)

delta: have a native implementation of _findsnapshot The function might traverse a lot of revision, a native implementation get significantly faster. example affected manifest write before: 0.114989 after: 0.067141 (-42%)

transaction: do not overwrite atomic-temp files on error Even though the original files can be restored from the backup, it should be better to not write back a temporary file if we know it can be corrupted.