Re: [PATCH v3 7/8] reset: make --mixed sparse-aware

From: Victoria Dye <vdye@github.com>
Date: Nov 22, 2021, 16:46 UTC
Message-ID: <c97e4252-b17a-c8c1-3bde-cbfe22a6e4d5@github.com>
In-Reply-To: <CABPp-BEcWutcLJWJvO-DC5B59EO4focgCJG1OWYg080j8ti44Q@mail.gmail.com>

Elijah Newren wrote:

Show 12 quoted lines

> On Thu, Oct 7, 2021 at 2:15 PM Victoria Dye via GitGitGadget
> <gitgitgadget@gmail.com> wrote:
>>
>> From: Victoria Dye <vdye@github.com>
>>
>> Remove the `ensure_full_index` guard on `read_from_tree` and update `git
>> reset --mixed` to ensure it can use sparse directory index entries wherever
>> possible. Sparse directory entries are reset use `diff_tree_oid`, which
> 
> I am having trouble parsing this second sentence.  Was this meant to
> be 'Sparse directory entries _which_ are reset use...'?
>

It should be "Sparse directory entries are reset _using_ `diff_tree_oid`...".

Show 116 quoted lines

>> requires `change` and `add_remove` functions to process the internal
>> contents of the sparse directory. The `recursive` diff option handles cases
>> in which `reset --mixed` must diff/merge files that are nested multiple
>> levels deep in a sparse directory.
>>
>> The use of pathspecs with `git reset --mixed` introduces scenarios in which
>> internal contents of sparse directories may be matched by the pathspec. In
>> order to reset *all* files in the repo that may match the pathspec, the
>> following conditions on the pathspec require index expansion before
>> performing the reset:
>>
>> * "magic" pathspecs
>> * wildcard pathspecs that do not match only in-cone files or entire sparse
>>   directories
>> * literal pathspecs matching something outside the sparse checkout
>>   definition
>>
>> Helped-by: Elijah Newren <newren@gmail.com>
>> Signed-off-by: Victoria Dye <vdye@github.com>
>> ---
>>  builtin/reset.c                          | 78 +++++++++++++++++++++++-
>>  t/t1092-sparse-checkout-compatibility.sh | 17 ++++++
>>  2 files changed, 93 insertions(+), 2 deletions(-)
>>
>> diff --git a/builtin/reset.c b/builtin/reset.c
>> index 0ac0de7dc97..60517e7e1d6 100644
>> --- a/builtin/reset.c
>> +++ b/builtin/reset.c
>> @@ -148,7 +148,9 @@ static void update_index_from_diff(struct diff_queue_struct *q,
>>                  * If the file 1) corresponds to an existing index entry with
>>                  * skip-worktree set, or 2) does not exist in the index but is
>>                  * outside the sparse checkout definition, add a skip-worktree bit
>> -                * to the new index entry.
>> +                * to the new index entry. Note that a sparse index will be expanded
>> +                * if this entry is outside the sparse cone - this is necessary
>> +                * to properly construct the reset sparse directory.
>>                  */
>>                 pos = cache_name_pos(one->path, strlen(one->path));
>>                 if ((pos >= 0 && ce_skip_worktree(active_cache[pos])) ||
>> @@ -166,6 +168,73 @@ static void update_index_from_diff(struct diff_queue_struct *q,
>>         }
>>  }
>>
>> +static int pathspec_needs_expanded_index(const struct pathspec *pathspec)
>> +{
>> +       unsigned int i, pos;
>> +       int res = 0;
>> +       char *skip_worktree_seen = NULL;
>> +
>> +       /*
>> +        * When using a magic pathspec, assume for the sake of simplicity that
>> +        * the index needs to be expanded to match all matchable files.
>> +        */
>> +       if (pathspec->magic)
>> +               return 1;
>> +
>> +       for (i = 0; i < pathspec->nr; i++) {
>> +               struct pathspec_item item = pathspec->items[i];
>> +
>> +               /*
>> +                * If the pathspec item has a wildcard, the index should be expanded
>> +                * if the pathspec has the possibility of matching a subset of entries inside
>> +                * of a sparse directory (but not the entire directory).
>> +                *
>> +                * If the pathspec item is a literal path, the index only needs to be expanded
>> +                * if a) the pathspec isn't in the sparse checkout cone (to make sure we don't
>> +                * expand for in-cone files) and b) it doesn't match any sparse directories
>> +                * (since we can reset whole sparse directories without expanding them).
>> +                */
>> +               if (item.nowildcard_len < item.len) {
>> +                       for (pos = 0; pos < active_nr; pos++) {
>> +                               struct cache_entry *ce = active_cache[pos];
>> +
>> +                               if (!S_ISSPARSEDIR(ce->ce_mode))
>> +                                       continue;
> 
> This double loop over all pathspecs and over all index entries reminds
> me of the original non-cone mode sparsity patterns.  Stolee introduced
> cone mode patterns specifically to avoid the expensiveness of such
> double loops (cf.
> https://lore.kernel.org/git/19d664a5dada87a9a8dcf18d7548582275593f10.1566313865.git.gitgitgadget@gmail.com/).
> Can one of the functions he added allow us to avoid this double loop,
> or are there complications that don't allow this (e.g. the actually
> SKIP_WORKTREE paths don't quite match the requested sparsity paths in
> some cases, or here we are faced with just a leading path of multiple
> index entries)?
> 
>> +                               /*
>> +                                * If the pre-wildcard length is longer than the sparse
>> +                                * directory name and the sparse directory is the first
>> +                                * component of the pathspec, need to expand the index.
>> +                                */
>> +                               if (item.nowildcard_len > ce_namelen(ce) &&
>> +                                   !strncmp(item.original, ce->name, ce_namelen(ce))) {
>> +                                       res = 1;
>> +                                       break;
>> +                               }
>> +
>> +                               /*
>> +                                * If the pre-wildcard length is shorter than the sparse
>> +                                * directory and the pathspec does not match the whole
>> +                                * directory, need to expand the index.
>> +                                */
>> +                               if (!strncmp(item.original, ce->name, item.nowildcard_len) &&
>> +                                   wildmatch(item.original, ce->name, 0)) {
>> +                                       res = 1;
>> +                                       break;
>> +                               }
>> +                       }
>> +               } else if (!path_in_cone_mode_sparse_checkout(item.original, &the_index) &&
>> +                          !matches_skip_worktree(pathspec, i, &skip_worktree_seen))
> 
> Oh, so you can at least generally avoid the double loop.  That's good.
> So is this just a case of wildcards are special and there isn't a way,
> even in cone-mode, to avoid the double loop?
>

The wildcard pathspecs are difficult to handle as "cleanly" as the non-wildcard pathspecs, since the condition for expanding the index is whether the pathspec has the potential to match some, but not all, contents of any given sparse directory. Luckily, the double loop is constrained to only sparse directories in the index and wildcard pathspecs, and exits as soon as there is any indication that the index needs to be expanded.

Show 6 quoted lines

> (Given that I'm so tardy in reviewing this, even if the answer is that
> the double loop is avoidable, or if we just don't know, I'd be totally
> fine with a 'TODO: consider whether this double loop could be avoided
> in cone mode using some kind of variant of
> path_in_cone_mode_sparse_checkout()')
>

I had originally tried something like this, but even if a wildcard pathspec is in-cone, it could match something outside of it. For example, '*.c' is in-cone, but has the potential to match only some of the files in a sparse directory containing both *.c and *.h files.

Looking at it again, though, if you constrain `path_in_cone_mode_sparse_checkout()` pathspecs further and require that the only post-`nowildcard_len` characters are '*', there's no risk of matching partial subsets of files in sparse directories. I'll add that in a re-roll.

Show 64 quoted lines

>> +                       res = 1;
>> +
>> +               if (res > 0)
>> +                       break;
>> +       }
>> +
>> +       free(skip_worktree_seen);
>> +       return res;
>> +}
>> +
>>  static int read_from_tree(const struct pathspec *pathspec,
>>                           struct object_id *tree_oid,
>>                           int intent_to_add)
>> @@ -178,9 +247,14 @@ static int read_from_tree(const struct pathspec *pathspec,
>>         opt.format_callback = update_index_from_diff;
>>         opt.format_callback_data = &intent_to_add;
>>         opt.flags.override_submodule_config = 1;
>> +       opt.flags.recursive = 1;
>>         opt.repo = the_repository;
>> +       opt.change = diff_change;
>> +       opt.add_remove = diff_addremove;
>> +
>> +       if (pathspec->nr && the_index.sparse_index && pathspec_needs_expanded_index(pathspec))
>> +               ensure_full_index(&the_index);
>>
>> -       ensure_full_index(&the_index);
>>         if (do_diff_cache(tree_oid, &opt))
>>                 return 1;
>>         diffcore_std(&opt);
>> diff --git a/t/t1092-sparse-checkout-compatibility.sh b/t/t1092-sparse-checkout-compatibility.sh
>> index 4ac93874cb2..c9343ff5b9c 100755
>> --- a/t/t1092-sparse-checkout-compatibility.sh
>> +++ b/t/t1092-sparse-checkout-compatibility.sh
>> @@ -774,11 +774,28 @@ test_expect_success 'sparse-index is not expanded' '
>>                 ensure_not_expanded reset --hard $ref || return 1
>>         done &&
>>
>> +       ensure_not_expanded reset --mixed base &&
>>         ensure_not_expanded reset --hard update-deep &&
>>         ensure_not_expanded reset --keep base &&
>>         ensure_not_expanded reset --merge update-deep &&
>>         ensure_not_expanded reset --hard &&
>>
>> +       ensure_not_expanded reset base -- deep/a &&
>> +       ensure_not_expanded reset base -- nonexistent-file &&
>> +       ensure_not_expanded reset deepest -- deep &&
>> +
>> +       # Although folder1 is outside the sparse definition, it exists as a
>> +       # directory entry in the index, so the pathspec will not force the
>> +       # index to be expanded.
>> +       ensure_not_expanded reset deepest -- folder1 &&
>> +       ensure_not_expanded reset deepest -- folder1/ &&
>> +
>> +       # Wildcard identifies only in-cone files, no index expansion
>> +       ensure_not_expanded reset deepest -- deep/\* &&
>> +
>> +       # Wildcard identifies only full sparse directories, no index expansion
>> +       ensure_not_expanded reset deepest -- folder\* &&
>> +
>>         ensure_not_expanded checkout -f update-deep &&
>>         test_config -C sparse-index pull.twohead ort &&
>>         (
>> --
>> gitgitgadget