Re: [PATCH 02/10] merge-recursive.c: call a new unpack_trees_options_init() function

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

On Mon, Oct 04 2021, Elijah Newren wrote:

> On Sun, Oct 3, 2021 at 5:46 PM Ævar Arnfjörð Bjarmason <avarab@gmail.com> wrote:
>>
>> In the preceding commit we introduced a new UNPACK_TREES_OPTIONS_INIT
>> macro, but "merge-recursive.c" could not be converted to it since
>> it (re-)initializes a "struct unpack_trees_options" on the heap.
>>
>> In order to convert use the macro as the source of truth for
>> initialization we need to not only convert the initialization in
>> unpack_trees_start(), but also call the new
>> unpack_trees_options_init() just after the CALLOC_ARRAY() in
>> merge_start().
>
> Um...why?

Replied below.

>> When we call merge_trees() we'll call merge_start() followed by
>> merge_trees_internal(), and it will call unpack_trees_start() before
>> it does much of anything. So it's covered by an initialization in
>> unpack_trees_start().
>>
>> But when merge_recursive() is called it will call merge_start()
>> followed by merge_recursive_internal(), which won't call
>> unpack_trees_start() until its own call call to
>> merge_trees_internal(), but in the meantime it might end up using that
>> calloc'd "struct unpack_trees_options".
>
> Nothing in merge-recursive.c usings unpack_opts before
> unpack_trees_start() or after unpack_trees_finish().  If anyone
> attempts to use it elsewhere, that would itself be a bug.  So, I'd
> replace the above three paragraphs with:
>
> "Change the initialization of opt->priv_unpack_opts from using memset
> to 0, with unpack_trees_options_init()."
>
> or something like that, and then drop your change to merge_start().

Sure, I'll defer to you about being confident about that. I didn't want
to leave a copy if it hanging without the proper initialization in case
there were new callers.

>> This was OK before, as setup_unpack_trees_porcelain() would call
>> strvec_init(), and our "struct dir_struct" in turn is OK with being
>> NULL'd. Let's convert the former to macro initialization, we'll deal
>> with the latter in a subsequent commit.
>
> This is quite confusing; it's really hard to understand how this
> relates to the rest of the commit message.  I have to read the code to
> see what you're doing, and then write my own commit message in my head
> because the wording in this paragraph still doesn't parse.
>
> I'd make the change strvec_init/STRVEC_INIT changes be a separate
> patch.  I suspect it'll be easier to write the commit message for it
> as a standalone commit as well.

Sure, FWIW I had it split up locally, and figured it would be easier to
explain if the API usage change came with the initialization change that
made it possible.

>>
>> Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
>> ---
>>  merge-recursive.c | 3 ++-
>>  unpack-trees.c    | 8 ++++++--
>>  unpack-trees.h    | 5 ++++-
>>  3 files changed, 12 insertions(+), 4 deletions(-)
>>
>> diff --git a/merge-recursive.c b/merge-recursive.c
>> index e594d4c3fa1..d24a4903f1d 100644
>> --- a/merge-recursive.c
>> +++ b/merge-recursive.c
>> @@ -405,7 +405,7 @@ static int unpack_trees_start(struct merge_options *opt,
>>         struct tree_desc t[3];
>>         struct index_state tmp_index = { NULL };
>>
>> -       memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
>> +       unpack_trees_options_init(&opt->priv->unpack_opts);
>>         if (opt->priv->call_depth)
>>                 opt->priv->unpack_opts.index_only = 1;
>>         else
>> @@ -3704,6 +3704,7 @@ static int merge_start(struct merge_options *opt, struct tree *head)
>>
>>         CALLOC_ARRAY(opt->priv, 1);
>>         string_list_init_dup(&opt->priv->df_conflict_file_set);
>> +       unpack_trees_options_init(&opt->priv->unpack_opts);
>
> Drop this hunk.

Speaking of splitting things up in more understandable ways: If we're
going to hard rely on the interaction between merge_start() and
unpack_trees_start() wouldn't it make sense to lead with a change that
dropped that memset, which if this invariant holds is also redundant to
the CALLOC() of opt->priv in merge_start() in the pre-image.