Christian Couder <christian.couder@xxxxxxxxx> writes:
> `git pack-objects` supports the `--missing=<missing-action>` option in
> the same way as `git rev-list` except when '<missing-action>' is
> "print", which `git pack-objects` doesn't support.
>
> As we want to refactor `git pack-objects` to use the same code from
> "missing.{c,h}" as `git rev-list` for the `--missing=...` feature, let's
> make it possible for that code to reject `--missing=print`.
>
> `git pack-objects` will then use that code in a following commit.
>
> Signed-off-by: Christian Couder <chriscool@xxxxxxxxxxxxx>
> ---
> builtin/rev-list.c | 2 +-
> missing.c | 4 ++--
> missing.h | 2 +-
> 3 files changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/builtin/rev-list.c b/builtin/rev-list.c
> index f71cc5ebe1..a712a6fd62 100644
> --- a/builtin/rev-list.c
> +++ b/builtin/rev-list.c
> @@ -539,7 +539,7 @@ int cmd_rev_list(int argc, const char **argv, const char *prefix)
> int res;
> if (revs.exclude_promisor_objects)
> die(_("options '%s' and '%s' cannot be used together"), "--exclude-promisor-objects", "--missing");
> - res = parse_missing_action_value(arg);
> + res = parse_missing_action_value(arg, 1);
Hmph, this smells like a horribly unscalable design, as we make the
vocabulary of missing-action richer, you'd end up piling on "this
choice is allowed in this call" parameters, wouldn't you? The first
person who adds such an ad-hoc parameter would say "hey, what's just
one extra parameter print_ok between friends", but the next person
would say the same thing for their new choice and adds frotz_ok, and
we'd be in chaos.
Rather, shouldn't the _caller_ decide if the parsed value is
something it does not like and barf?
Alternatively, add a _single_ "reject" bitset and do something like
int parse_missing_action_value(const char *value, unsigned reject)
{
...
if (!(reject & (1<<MA_ERROR) && !strcmp(value, "error")))
return MA_ERROR;
if (!(reject & (1<<MA_PRINT) && !strcmp(value, "print")))
return MA_PRINT;
...
which would scale better (but still my preference is to have the
caller deal with only the values it recognises---do not make the
caller say "if (res >= 0 && res != MA_PRINT)" as that will not scale
when new choices that are accepted elsewhere are added.
