I just have nowhere near enough CTF knowledge to even try to comment.
but if it makes cpu_t print, I'm all for it. Style-wise, it doesn't look
ugly at all to me.
John Levon wrote:
> On Sat, Nov 04, 2006 at 06:38:26PM +0000, John Levon wrote:
>
>> In theory it's not hard to fix, I believe we'd have to make
>> mdb_ctf_type_resolve() iterate through all the ctf's for each module.
>
> Looking a little harder, that won't do. The main problem is that ::print uses
> ctf_type_visit(); since this can't know about our other CTF containers, it
> won't be able to visit the cpu_m node usefully.
>
> It won't win me any awards for style, but the patch below at least allows
> ::print cpu_t cpu_m to work. An alternative would be to re-implement
> ctf_type_visit() in MDB, unless someone else has a good suggestion...
>
> regards
> john
>
>
> ------- usr/src/cmd/mdb/common/mdb/mdb_print.c -------
>
> Index: usr/src/cmd/mdb/common/mdb/mdb_print.c
> --- /ws/onnv-clone/usr/src/cmd/mdb/common/mdb/mdb_print.c Thu Aug 24
> 23:02:24 2006
> +++ /builds/johnlev/onnv-johnlev/usr/src/cmd/mdb/common/mdb/mdb_print.c
> Mon Nov 6 12:51:58 2006
> @@ -1754,6 +1754,26 @@
> }
>
> /*
> + * If we only resolved as far as a forward declaration, try to look it up in
> + * the other target objects, so ::print cpu_t cpu_m does something useful.
> + */
> +static void
> +resolve_forward(mdb_ctf_id_t *idp, int *kindp)
> +{
> + char buf[MDB_SYM_NAMLEN];
> + mdb_ctf_id_t fwd;
> +
> + if (mdb_ctf_type_name(*idp, buf, MDB_SYM_NAMLEN) == NULL)
> + return;
> +
> + if (mdb_ctf_lookup_by_name(buf, &fwd) != 0)
> + return;
> +
> + *idp = fwd;
> + *kindp = mdb_ctf_type_kind(fwd);
> +}
> +
> +/*
> * Recursively descend a print a given data structure. We create a struct of
> * the relevant print arguments and then call mdb_ctf_type_visit() to do the
> * traversal, using elt_print() as the callback for each element.
> @@ -1909,6 +1929,12 @@
> goto out;
> }
>
> + (void) mdb_ctf_type_resolve(mid, &rid);
> + kind = mdb_ctf_type_kind(rid);
> +
> + if (kind == CTF_K_FORWARD)
> + resolve_forward(&mid, &kind);
> +
> /*
> * If the member string ends with a "[0]"
> * (last_deref * is true) and the type is a
> @@ -1915,8 +1941,6 @@
> * structure or union, * print "->" rather
> * than "[0]." in elt_print.
> */
> - (void) mdb_ctf_type_resolve(mid, &rid);
> - kind = mdb_ctf_type_kind(rid);
> if (last_deref && IS_SOU(kind)) {
> char *end;
> (void) mdb_snprintf(buf, sizeof (buf),
