Re: [PATCH 0/5] Start of a journey: drop NO_THE_INDEX_COMPATIBILITY_MACROS

From: Junio C Hamano <gitster@pobox.com>
Date: May 5, 2017, 17:22 UTC
Message-ID: <xmqq8tmb8bhu.fsf@gitster.mtv.corp.google.com>
In-Reply-To: <20170504191927.GB15203@aiede.svl.corp.google.com>

Jonathan Nieder <jrnieder@gmail.com> writes:

> That is, one way to do what this series attempts would be the
> following:
>
>  1. rename variables that shadow the_index.

No question about this one.  It is a good thing to do.

>  2. add coccinelle patches (or one coccinelle patch) to
>     contrib/coccinelle implementing *_cache -> *_index migration.
>     Is there a way to do this without making it fail "make coccicheck"?

Quite honestly, I do not see much value in this, but take it merely as my knee-jerk reaction. The only scenario I can think of in which dropping *_cache() macros is an improvement as the end result is when our goal is to completely drop the singleton index_state instance, aka "the_index". I actually think that it may be a worthwhile goal to eradicate "the_index".

I wonder if somebody can take a small example codepath and make it not to rely on the existence of "the_index" from start to end? Have an instance of index_state on the stack of cmd_foo(), have it call read_index() into it where it currently calls read_cache(), update the support functions it calls so that it can pass the pointer to its index_info throughout the callchain, and see how involved the necessary changes of all of the above are. Start from something simple and small, e.g. "ls-files". The infrastructure code updated for such an experiment may be NO_THE_INDEX_COMPATIBILITY_MACROS clean.

Perhaps we can remove the existence of the_index from the system by going that route; needless to say, when that goal is achieved, by definition *_cache() macros will be completely useless and must be removed, as there is nobody who relies on the existence of and who uses "the_index".

>  3. migrate library code (but not builtins) using that semantic patch

I do think this is going backwards. The only thinng replacing *_cache() to *_index(&the_index) buys you is newbies not having to know both, but instead they will be exposed to the pattern of repeated use of *_index(&the_index); i.e. reliance of the existence of the singleton "the_index" is not reduced, but is stressed which is a big downside when we are trying to eventually get rid of it.

Also quite honestly, I do not think we want newbies to be touching the things that needs *_index() interface while this update is going on.

The remainder of your enumeration goes in a direction different from getting rid of the_index, so I do not know---I wrote the above under the assumption that the total removal of "the_index" might be a good thing to do, and where that assumption would lead us to (e.g. an obvious side-effect of no longer having the_index is that *_cache() macros cannot exist), and I am undecided if the assumption is a good one (yet).