Re: [PATCH 2/2] sparse-index.c: don't call prepare_repo_settings() twice in a row

[Derrick Stolee <stolee@gmail.com>]

On 5/6/2021 4:49 AM, Ævar Arnfjörð Bjarmason wrote:
> 
> On Wed, May 05 2021, Derrick Stolee wrote:
> 
>> On 5/5/2021 8:11 AM, Ævar Arnfjörð Bjarmason wrote:
>>> Change code added in 58300f47432 (sparse-index: add index.sparse
>>> config option, 2021-03-30) to only call prepare_repo_settings()
>>> once. We know that our own set_sparse_index_config() has just finished
>>> calling it, so we don't need to call it if we're acting on "test_env".
>>
>> I'm not sure about the value here. prepare_repo_settings() returns
>> quickly if the settings have already been prepared, so a second call
>> is negligible in cost.
> 
> I changed that while I was at it to make it easier to read, it's not an
> optimization. I.e. one wonders what the side-effect is of calling
> prepare_repo_settings() twice, discovers there isn't one...

This is typical of the "prepare_" pattern, such as prepare_packed_git()
or prepare_commit_graph(). It's saying "be sure this is initialized"
and that initialization only needs to happen once.
 
>>> @@ -133,11 +133,12 @@ int convert_to_sparse(struct index_state *istate)
>>>  	test_env = git_env_bool("GIT_TEST_SPARSE_INDEX", -1);
>>>  	if (test_env >= 0)
>>>  		set_sparse_index_config(istate->repo, test_env);
>>> +	else
>>> +		prepare_repo_settings(istate->repo);
>>
>> The change presented here to either call set_sparse_index_config()
>> _or_ prepare_repo_settings() seems like it knows too much about
>> how set_sparse_index_config() works.
> 
> It seems reasonable to assume that a function to set config has
> initialized (or died, if it couldn't) enough of our config state to do
> its job.
> 
> Besides, it's a few lines above the changed code in the same file. 
> 
> But looking at this again 2/3 callers of set_sparse_index_config()
> aren't checking the return value. Wouldn't something like [1] on top be
> a good idea here?
> 
> Also, is GIT_TEST_SPARSE_INDEX=true itself supposed to work? Running the
> test suite with it fails 3 test files for me, all /sparse/, i.e. tests
> that (presumably) assume it's not already turned on by this code.

I haven't ensured that this works in all cases. I was going to try and
find a way to make the tests be better about having a meaningful data
shape that applies to the sparse-index, but I should go and verify that
the tests work with this enabled.

> 
> 1. 
> 
> diff --git a/builtin/sparse-checkout.c b/builtin/sparse-checkout.c
> index a4bdd7c4940..bea1e7dd00e 100644
> --- a/builtin/sparse-checkout.c
> +++ b/builtin/sparse-checkout.c
> @@ -280,8 +280,9 @@ static int set_config(enum sparse_checkout_mode mode)
>  				      "core.sparseCheckoutCone",
>  				      mode == MODE_CONE_PATTERNS ? "true" : NULL);
>  
> -	if (mode == MODE_NO_PATTERNS)
> -		set_sparse_index_config(the_repository, 0);
> +	if (mode == MODE_NO_PATTERNS &&
> +	    set_sparse_index_config(the_repository, 0) < 0)
> +		die(_("could not set index.sparse=0"));

This seems reasonable, because we are in a builtin saying "please disable
the sparse-index". Is there not an instance here that _enables_ it?

I would change the string to be "could not disable index.sparse" or
something similar.

>  	return 0;
>  }
> diff --git a/sparse-index.c b/sparse-index.c
> index 5bad05de645..3938bcec962 100644
> --- a/sparse-index.c
> +++ b/sparse-index.c
> @@ -131,10 +131,13 @@ int convert_to_sparse(struct index_state *istate)
>  	 * index.sparse config variable to be on.
>  	 */
>  	test_env = git_env_bool("GIT_TEST_SPARSE_INDEX", -1);
> -	if (test_env >= 0)
> -		set_sparse_index_config(istate->repo, test_env);
> -	else
> +	if (test_env >= 0) {
> +		if (set_sparse_index_config(istate->repo, test_env) < 0)
> +			die(_("could not set index.sparse based on GIT_TEST_SPARSE_INDEX=%d"),
> +			    test_env);
> +	} else {
>  		prepare_repo_settings(istate->repo);
> +	}

This one, I'm not so sure. There might be reasons why the GIT_TEST_*
variable won't work to set the config and we don't want that to be
a reason the test fails. There is no need for a translated string
here, either.

Thanks,
-Stolee