Simon Sapin <simon.sapin@octobus.net> [Thu, 14 Oct 2021 16:39:16 +0200] rev 48232
dirstate-v2: Separate HAS_FILE_MTIME and HAS_DIRECTORY_MTIME flags
Previously the same flag was used, with its meaning based on whether the node
otherwise identifies a file tracked anywhere.
In addition to being more explicit, this enables storing a directory mtime
if a given path used to be tracked in a parent commit (so the dirstate still
has data about it) but became a directory in the working copy.
(However this is not done yet as it would require a larger change,
replacing the `dirstate_map::NodeData` enum with struct fields.)
Differential Revision: https://phab.mercurial-scm.org/D11662
Simon Sapin <simon.sapin@octobus.net> [Thu, 14 Oct 2021 16:06:31 +0200] rev 48231
dirstate-v2: Extend node flags to 16 bits
Only 7 out of 8 available bits are used right now. Reserve some more.
Future versions of Mercurial may assign meaning to some of these bits,
with the limitation that then-older versions will always reset those bits to
unset when writing nodes.
(A new node is written for any mutation in its subtree, leaving the bytes of
the old node unreachable until the data file is rewritten entirely.)
Differential Revision: https://phab.mercurial-scm.org/D11661
Martin von Zweigbergk <martinvonz@google.com> [Thu, 14 Oct 2021 08:58:07 -0700] rev 48230
narrow: raise StateError when working copy is stale (for detailed exit code)
Differential Revision: https://phab.mercurial-scm.org/D11663
Simon Sapin <simon.sapin@octobus.net> [Thu, 14 Oct 2021 15:05:04 +0200] rev 48229
dirstate-v2: Use attributes as intended instead of properties in v2_data()
The property return other integer values instead of None, so `is not None`
does not work.
This fixes test-dirstate-race.t in pure-Python mode, which currently fails
on the default branch.
Differential Revision: https://phab.mercurial-scm.org/D11660
Simon Sapin <simon.sapin@octobus.net> [Thu, 14 Oct 2021 13:59:18 +0200] rev 48228
dirstate: Remove unused variable
This fixes test-check-pyflakes.t which is currently failing
on the default branch.
Differential Revision: https://phab.mercurial-scm.org/D11658
Simon Sapin <simon.sapin@octobus.net> [Thu, 14 Oct 2021 13:58:25 +0200] rev 48227
rust: Reformat source code
This fixes test-check-rust-format.t which is currently failing
on the default branch.
Differential Revision: https://phab.mercurial-scm.org/D11657
Arseniy Alekseyev <aalekseyev@janestreet.com> [Thu, 14 Oct 2021 13:34:37 +0100] rev 48226
rhg: do not try to open a nodemap for an inline index
This saves an [open] system call per file, which is a small saving, but
it showed up in the profile at large file counts (it accounted for 30ms
out of 400ms needed for catting 10000 files, on a ZFS filesystem on Linux,
so ~3us per syscall).
Differential Revision: https://phab.mercurial-scm.org/D11659
Arseniy Alekseyev <aalekseyev@janestreet.com> [Tue, 05 Oct 2021 15:10:42 +0100] rev 48225
rhg: stop manifest traversal when no more files are needed
Stopping the traversal early can skip a significant part
of the manifest traversal, to avoid some of its cost.
The worst-case benchmarks are favorable, as well.
Running [hg cat] on the last file in the manifest of
a large repo, I'm seeing a ~4ms improvement (150ms -> 146ms),
so this time is now almost indistinguishable from the
baseline ("brute force") implementation.
Running [hg cat] on ~220 files together with the last file
of the repo is further improved by ~5ms or so.
I suspect the raw performance improvements are caused by splitting
the manifest search and the file data access into separate phases,
instead of interleaving them.
Differential Revision: https://phab.mercurial-scm.org/D11616
Arseniy Alekseyev <aalekseyev@janestreet.com> [Mon, 04 Oct 2021 19:06:45 +0100] rev 48224
rhg: faster hg cat when many files are requested
With this patch I'm seeing a ~39ms improvement (220ms -> 181ms) when
running [hg cat] on ~220 files in a ~260k-file repo.
The timing for [hg cat] on an individual file becomes slightly worse
(losing 5ms: 145ms -> 150ms).
A follow-up commit is intended to improve that.
Differential Revision: https://phab.mercurial-scm.org/D11615
Simon Sapin <simon.sapin@octobus.net> [Fri, 16 Jul 2021 18:42:20 +0200] rev 48223
dirstate-v2: Add support when Rust is not enabled
This wires into `dirstatemap` the parser and serializer added in previous
changesets. The memory representation is still the same, with a flat `dict`
for `DirstateItem`s and another one for copy sources. Serialization always
creates a new dirstate-v2 data file and does not support (when Rust is not
enabled) appending to an existing one, since we don’t keep track of which
tree nodes are new or modified. Instead the tree is reconstructed during
serialization.
Differential Revision: https://phab.mercurial-scm.org/D11520