Re: [PATCH 0/5] *.[ch]: don't duplicate *_init() and *_INIT logic

[Jeff King <peff@peff.net>]

On Thu, Jul 01, 2021 at 12:51:24PM +0200, Ævar Arnfjörð Bjarmason wrote:

> This series implements a suggestion by Jeff King to use an idiom for
> *_init() functions that avoids duplicating what we've declared in the
> corresponding *_INIT macros. See
> https://lore.kernel.org/git/YNytp0JAIaQih0Y4@coredump.intra.peff.net/

These all look pretty good to me. My comments are minor enough that I'll
include them here rather than individual replies:

> Ævar Arnfjörð Bjarmason (5):
>   *.h: move some *_INIT to designated initializers

I agree with Martin's suggestion to prefer the designated initializer
for STRING_LIST_INIT_NODUP. The consistency makes it easier to read.

>   *.c *_init(): define in terms of corresponding *_INIT macro

Yep, nice.

>   dir.[ch]: replace dir_init() with DIR_INIT

I like this. Since there is a dir_clear(), I guess somebody could argue
that lacking an init function makes things less consistent/balanced. But
if we don't need it, I'm happy to drop it.

>   string-list.[ch]: add a string_list_init_{nodup,dup}()

I like this. I agree the current "0/1" in the callers is kind of
inscrutable. It does make things harder if we wanted to add more
options, but that is already true due to the macro initializers.

>   string-list.h users: change to use *_{nodup,dup}()

Nice. IMHO you should drop the convenience wrapper here. Leaving it
means we'll have to circle back later and fix up topics in flight (and
it's not atomic; when we finally drop it, there may be more topics in
flight then!).

If it were somehow complicated to convert callers, that might be more
justified. But doing a quick conversion (which is easily detected by the
compiler) in the merge commit seems like less work than having to
revisit the topic later.

-Peff