Re: [PATCH v2 0/5] Use complete_action's todo list to do the rebase

[Johannes Schindelin <Johannes.Schindelin@gmx.de>]

[-- Attachment #1: Type: text/plain, Size: 6540 bytes --]

Hi Alban,

On Mon, 7 Oct 2019, Alban Gruin wrote:

> This can be seen as a continuation of ag/reduce-rewriting-todo.
>
> Currently, complete_action() releases its todo list before calling
> sequencer_continue(), which reloads the todo list from the disk.  This
> series removes this useless round trip.
>
> Patches 1, 2, and 3 originally come from a series meaning to improve
> rebase.missingCommitsCheck[0].  In the original series, I wanted to
> check for missing commits in read_populate_todo(), so a warning could be
> issued after a `rebase --continue' or an `exec' commands.  But, in the
> case of the initial edit, it is already checked in complete_action(),
> and would be checked a second time in sequencer_continue() (a caller of
> read_populate_todo()).  So I hacked up sequencer_continue() to accept a
> pointer to a todo list, and if not null, would skip the call to
> read_populate_todo().  (This was really ugly, to be honest.)  Some
> issues arose with git-prompt.sh[1], hence 1, 2 and 3.
>
> Patch 5 is a new approach to what I did first.  Instead of bolting a new
> parameter to sequencer_continue(), this makes complete_action() calling
> directly pick_commits().
>
> This is based on 4c86140027 ("Third batch").
>
> Changes since v1:
>
>  - Rewording of patches 1, 2, 4 and 5 according to comments made by
>    Phillip Wood, Junio C Hamano and Johannes Schindelin.
>
> The tip of this series is tagged as "reduce-todo-list-cont-v2" at
> https://github.com/agrn/git.
>
> [0] http://public-inbox.org/git/20190717143918.7406-1-alban.gruin@gmail.com/
> [1] http://public-inbox.org/git/1732521.CJWHkCQAay@andromeda/
>
> Alban Gruin (5):
>   sequencer: update `total_nr' when adding an item to a todo list
>   sequencer: update `done_nr' when skipping commands in a todo list
>   sequencer: move the code writing total_nr on the disk to a new
>     function
>   rebase: fill `squash_onto' in get_replay_opts()
>   sequencer: directly call pick_commits() from complete_action()
>
>  builtin/rebase.c |  5 +++++
>  sequencer.c      | 26 ++++++++++++++++++--------
>  2 files changed, 23 insertions(+), 8 deletions(-)
>
> Diff-intervalle contre v1 :
> 1:  d177b0de1a ! 1:  9215b191c7 sequencer: update `total_nr' when adding an item to a todo list
>     @@ Metadata
>       ## Commit message ##
>          sequencer: update `total_nr' when adding an item to a todo list
>
>     -    `total_nr' is the total amount of items, done and toto, that are in a
>     -    todo list.  But unlike `nr', it was not updated when an item was
>     -    appended to the list.
>     +    `total_nr' is the total number of items, counting both done and todo,
>     +    that are in a todo list.  But unlike `nr', it was not updated when an
>     +    item was appended to the list.
>
>          This variable is mostly used by command prompts (ie. git-prompt.sh and
>     -    the like).
>     +    the like).  By forgetting to update it, the original code made it not
>     +    reflect the reality, but this flaw was masked by the code calling
>     +    unnecessarily read_todo_list() again to update the variable to its
>     +    correct value.  At the end of this series, the unnecessary call will be
>     +    removed, and the inconsistency addressed by this patch would start to
>     +    matter.
>
>          Signed-off-by: Alban Gruin <alban.gruin@gmail.com>
>
> 2:  09fcbe159b ! 2:  7cad541092 sequencer: update `done_nr' when skipping commands in a todo list
>     @@ Commit message
>          or skipped, but skip_unnecessary_picks() did not update it.
>
>          This variable is mostly used by command prompts (ie. git-prompt.sh and
>     -    the like).
>     +    the like).  As in the previous commit, this inconsistent behaviour is
>     +    not a problem yet, but it would start to matter at the end of this
>     +    series the same reason.
>
>          Signed-off-by: Alban Gruin <alban.gruin@gmail.com>
>
> 3:  26a18cd1a9 = 3:  7c9c4ddd30 sequencer: move the code writing total_nr on the disk to a new function
> 4:  5d74903cfe ! 4:  cd44fb4e10 rebase: fill `squash_onto' in get_replay_opts()
>     @@ Metadata
>       ## Commit message ##
>          rebase: fill `squash_onto' in get_replay_opts()
>
>     -    get_replay_opts() did not fill `squash_onto' if possible, meaning that
>     -    this field should be read from the disk by the sequencer through
>     -    read_populate_opts().  Without this, calling `pick_commits()' directly
>     -    will result in incorrect results with `rebase --root'.
>     +    Currently, the get_replay_opts() function does not initialise the
>     +    `squash_onto' field (which is used for the `--root' mode), only
>     +    read_populate_opts() does.  That would lead to incorrect results when
>     +    calling pick_commits() without reading the options from the disk first.
>
>          Let’s change that.
>
> 5:  dc803c671f ! 5:  523fdd35a1 sequencer: directly call pick_commits() from complete_action()
>     @@ Commit message
>          sequencer: directly call pick_commits() from complete_action()
>
>          Currently, complete_action() calls sequencer_continue() to do the
>     -    rebase.  Even though the former already has the todo list, the latter
>     -    loads it from the disk and parses it.  Calling directly pick_commits()
>     -    from complete_action() avoids this unnecessary round trip.
>     +    rebase.  Before the former calls pick_commits(), it
>     +
>     +     - calls read_and_refresh_cache() -- this is unnecessary here as we've
>     +       just called require_clean_work_tree()
>     +     - calls read_populate_opts() -- this is unnecessary as we're starting a
>     +       new rebase, so opts is fully populated
>     +     - loads the todo list -- this is unnecessary as we've just populated
>     +       the todo list
>     +     - commits any staged changes -- this is unnecessary as we're starting a
>     +       new rebase, so there are no staged changes
>     +     - calls record_in_rewritten() -- this is unnecessary as we're starting
>     +       a new rebase.
>     +
>     +    This changes complete_action() to directly call pick_commits() to avoid
>     +    these unnecessary steps.
>
>          Signed-off-by: Alban Gruin <alban.gruin@gmail.com>

This range-diff looks good to me!

I just verified that b2 addresses all the concerns I offered for v1.

Thanks,
Dscho