On Wed, Sep 28, 2016 at 07:01:38AM +0200, Michael Haggerty wrote:

> >   - a global for chaining to error, like:
> > 
> >        struct error_context print_errors = {
> >           error, /* actually a wrapper to handle va_list and NULL data */
> >           NULL
> >        };
> There could also be a global for chaining to `warn()` or `die()`.

I played around a little with this. The latter actually makes a lot of
code cleaner, because we can rely on the functions not returning at all.
So for example, you get:

diff --git a/branch.c b/branch.c
index a5a8dcb..53404b8 100644
--- a/branch.c
+++ b/branch.c
@@ -303,17 +303,13 @@ void create_branch(const char *head,
        if (!dont_change_ref) {
                struct ref_transaction *transaction;
-               struct strbuf err = STRBUF_INIT;
-               transaction = ref_transaction_begin(&err);
-               if (!transaction ||
-                   ref_transaction_update(transaction, ref.buf,
-                                          sha1, forcing ? NULL : null_sha1,
-                                          0, msg, &err) ||
-                   ref_transaction_commit(transaction, &err))
-                       die("%s", err.buf);
+               transaction = ref_transaction_begin(&error_die);
+               ref_transaction_update(transaction, ref.buf,
+                                      sha1, forcing ? NULL : null_sha1,
+                                      0, msg, &error_die);
+               ref_transaction_commit(transaction, &error_die);
-               strbuf_release(&err);
        if (real_ref && track)

which is much shorter and to the point (it does rely on the called
functions always calling report_error() and never just returning NULL or
"-1", but that should be the already. If it isn't, we'd be printing
"fatal: " with no message).

Cases that call:

  error("%s", err.buf);

can drop the strbuf handling, but of course still need to retain their
conditionals. So they're better, but not as much. I did a half-hearted
conversion of some of the ref code that uses strbufs, and it seems like
it would save a few hundred lines of boilerplate.

There are some cases that are _worse_, because they want to prefix the
error. E.g., in init-db, we have:

  struct strbuf err = STRBUF_INIT;
  if (refs_init_db(&err))
        die("failed to set up refs db: %s", err.buf);

which is fairly clean. Using an error_context adds slightly to the

  struct strbuf err_buf = STRBUF_INIT;
  struct error_context err = STRBUF_ERR(&err_buf);
  if (refs_init_db(&err))
        die("failed to set up refs db: %s", err_buf.buf);

Though if we wanted to get really magical, the err_buf/err pattern could
be its own single-line macro.

You could solve this more generally with something like:

  struct error_prefix_data err;

  error_prefix(&err, &error_die, "failed to set up refs db");

where error_prefix() basically sets us up to call back a function which
concatenates the prefix to the real error, then chains to error_die.
But to cover all cases, error_prefix() would actually have to format the
prefix string. Because some callers would be more like:

  error_prefix(&err, &error_print, "unable to frob %s", foo);
  do_frob(foo, &err);

We can't just save the va_list passed to error_prefix(), because it's
not valid after we return. So you have to format the prefix into a
buffer, even though in most cases we won't see an error at all (and
doing it completely correctly would involve using a strbuf, which means
there needs to be a cleanup step; yuck).


Reply via email to