Re: [PATCH v2 2/6] builtin/am: make sure state files are text

Jeff King [off-list ref] writes:

On Tue, Aug 25, 2015 at 09:19:13AM -0700, Junio C Hamano wrote:

quoted

As to "flags exposed to callers" vs "with and without gently", when
we change the system to allow new modes of operations (e.g. somebody
wants to write a binary file, or allocate more flag bits for their
special case), I'd expect that we'd add a more general and verbose
"write_file_with_options(path, flags, fmt, ...)"), gain experience
with that function, and then possibly introduce canned thin wrappers
(e.g. write_binary_file() that is a synonym to passing BINARY but
not GENTLY) if the new thing proves widely useful, just like I left
write_file() and write_file_gently() in as fairly common things to
do.

Yeah, that works. It is a bit of a gamble to me. If we never add a lot
more options, the end result is much nicer (callers do not deal with the
flag option at all). But if we do, we end up with the mess that
get_sha1_with_* and add_pending_object() got into.

Yeah.  I do not know.  Perhaps a good intermim solution for now
would be to make

  - write_file_l(path, flags, fmt, ...);

the low-level helper, with a single convenience wrapper:

  - write_file(path, fmt, ...)

for everybody other than two "gently" ones to use.  Two "gently"
ones can call write_file_l(path, WRITE_FILE_GENTLY, fmt,...).

You are right about binary stuff.  You could do fmt="...%c..."  and
pass '\0' to corresponding place if your NUL is in the fixed part of
the data you are generating, but otherwise write_file() interface is
a very useful way to handle binary.