Re: What's cooking in git.git (Jun 2021, #07; Wed, 30)

["Ævar Arnfjörð Bjarmason" <avarab@gmail.com>]

On Wed, Jun 30 2021, Junio C Hamano wrote:

> * jh/builtin-fsmonitor (2021-05-24) 30 commits
>  - t/perf: avoid copying builtin fsmonitor files into test repo
>  - t7527: test status with untracked-cache and fsmonitor--daemon
>  - p7519: add fsmonitor--daemon
>  - t7527: create test for fsmonitor--daemon
>  - fsmonitor: force update index after large responses
>  - fsmonitor: enhance existing comments
>  - fsmonitor--daemon: use a cookie file to sync with file system
>  - fsmonitor--daemon: periodically truncate list of modified files
>  - fsmonitor--daemon: implement handle_client callback
>  - fsmonitor-fs-listen-macos: implement FSEvent listener on MacOS
>  - fsmonitor-fs-listen-macos: add macos header files for FSEvent
>  - fsmonitor-fs-listen-win32: implement FSMonitor backend on Windows
>  - fsmonitor--daemon: create token-based changed path cache
>  - fsmonitor--daemon: define token-ids
>  - fsmonitor--daemon: add pathname classification
>  - fsmonitor--daemon: implement daemon command options
>  - fsmonitor-fs-listen-macos: stub in backend for MacOS
>  - fsmonitor-fs-listen-win32: stub in backend for Windows
>  - t/helper/fsmonitor-client: create IPC client to talk to FSMonitor Daemon
>  - fsmonitor--daemon: implement client command options
>  - fsmonitor--daemon: add a built-in fsmonitor daemon
>  - fsmonitor: introduce `core.useBuiltinFSMonitor` to call the daemon via IPC
>  - config: FSMonitor is repository-specific
>  - help: include fsmonitor--daemon feature flag in version info
>  - fsmonitor-ipc: create client routines for git-fsmonitor--daemon
>  - fsmonitor--daemon: update fsmonitor documentation
>  - fsmonitor--daemon: man page
>  - simple-ipc: preparations for supporting binary messages.
>  - Merge branch 'jk/perf-in-worktrees' into HEAD
>  - Merge branch 'jh/simple-ipc' into jh/rfc-builtin-fsmonitor
>
>  An attempt to write and ship with a watchman equivalent tailored
>  for our use.
>
>  What's the status of this one?

I think Johannes's reply to the last WC applies[1]:

    I am not Jeff, but I know that he is busy getting back to it, and
    plans on submitting a third iteration.

FWIW I'm still curious about some details on the performance concerns
that seem to have prompted this built-in fsmonitor endeavor, as I asked
about (but didn't get a reply to) in [2].

Not as a "we shouldn't have this, let's keep using the hook", but just
curiosity about why we've seemingly gotten such different performance
numbers on the watchman hook v.s. a built-in approach.

I suspect (but don't know) that the reason is that the built-in is
perhaps integrating differently with git somehow, in a way that we could
retrofit the hook approach to also do (if anyone still cares about the
hook approach).

In any case I'm interested in *why* the new approach is faster, given
that I've done some testing (again, noted in [2]) that suggest that
bottleneck in the previous pipeline wasn't at all what Jeff H. thought
it was.

1. https://lore.kernel.org/git/nycvar.QRO.7.76.6.2106171135530.57@tvgsbejvaqbjf.bet/#t
2. https://lore.kernel.org/git/87h7lgfchm.fsf@evledraar.gmail.com/