Re: [PATCH v2 3/4] bisect--helper: reimplement `bisect_run` shell function in C

El jue, 8 abr 2021 a las 0:09, Junio C Hamano ([off-list ref]) escribió:

Miriam Rubio [off-list ref] writes:

quoted

From: Tanushree Tumane <redacted>

Reimplement the `bisect_run()` shell function
in C and also add `--bisect-run` subcommand to
`git bisect--helper` to call it from git-bisect.sh.

Mentored-by: Christian Couder [off-list ref]
Signed-off-by: Tanushree Tumane <redacted>
Signed-off-by: Miriam Rubio <redacted>

If I am reading the patch correctly, this removes the need for the
$GIT_DIR/BISECT_RUN file that used to be used to keep track of the
state?  If that is true, it is worth noting in the proposed log
message.

As far as I can see, nobody creates $GIT_DIR/BISECT_RUN anymore.

    $ git grep -e path_bisect_run -e BISECT_RUN
    bisect.c:static GIT_PATH_FUNC(git_path_bisect_run, "BISECT_RUN")
    bisect.c:   unlink_or_warn(git_path_bisect_run());
    builtin/bisect--helper.c:           BISECT_RUN,
    builtin/bisect--helper.c:                    N_("use <cmd>... to automatical...
    builtin/bisect--helper.c:   case BISECT_RUN:
    t/t6030-bisect-porcelain.sh:        test_path_is_missing ".git/BISECT_RUN" &&

What if a run script tried to read from (or checked the presence of)
the file for its correct operation (e.g. I would imagine that "do
this operation when run interactively, but do the same operation
silently when run from the git-bisect machinery" may be a reasonable
thing to do)?

This change just unintendedly broke such a script, didn't it?  The
change makes me a bit worried.

Hi,
thank you for reviewing!.
I don't know why the need for the $GIT_DIR/BISECT_RUN file was
removed, so in the last patch series version I have just sent,
I have added the creation of the file and it contains bisect_state()
output as in original shell script version.
Regards,
Miriam

quoted

+     if (bisect_next_check(terms, NULL))
+             return BISECT_FAILED;
+
+     if (argc)
+             sq_quote_argv(&command, argv);
+     else
+             return BISECT_FAILED;
+
+     run_args.v[0] = xstrdup(command.buf);
+     run_args.nr = 1;
+
+     while (1) {
+             strvec_clear(&args);
+
+             printf(_("running %s"), command.buf);
+             res = run_command_v_opt(run_args.v, RUN_USING_SHELL);

Nicely used sq_quote_argv() with RUN_USING_SHELL here.  Goodl.

quoted

+             if (res < 0 && res >= 128) {
+                     error(_("bisect run failed: exit code %d from"
+                             " '%s' is < 0 or >= 128"), res, command.buf);
+                     strbuf_release(&command);
+                     return res;
+             }
+
+             if (res == 125)
+                     strvec_push(&args, "skip");
+             else if (res > 0)
+                     strvec_push(&args, terms->term_bad);
+             else
+                     strvec_push(&args, terms->term_good);
+

bisect_state() does so much that it was a bit hard to follow for me
(who hasn't been following the bisect-in-C topic very closely), but
the code around here roughly corresponds to the following snippet in
the original scripted version.

quoted

-             git bisect--helper --bisect-state $state >"$GIT_DIR/BISECT_RUN"
-             res=$?
-
-             cat "$GIT_DIR/BISECT_RUN"
-
-             if sane_grep "first $TERM_BAD commit could be any of" "$GIT_DIR/BISECT_RUN" \
-                     >/dev/null
-             then
-                     gettextln "bisect run cannot continue any more" >&2
-                     exit $res
-             fi

I see that the contents of the file BISECT_RUN is shown to the user
in the original but is that part of what bisect_state() does, or did
we lose it during this round of conversion?

quoted

+             res = bisect_state(terms, args.v, args.nr);
+             if (res == BISECT_INTERNAL_SUCCESS_MERGE_BASE) {
+                     printf(_("bisect run success"));
+                     res = BISECT_OK;
+             } else if (res == BISECT_ONLY_SKIPPED_LEFT)
+                     error(_("bisect run cannot continue any more"));
+             else if (res)
+                     error(_("bisect run failed:'git bisect--helper --bisect-state"
+                             " %s' exited with error code %d"), args.v[0], res);
+             else
+                     continue;

In any case, being able to check the return value from bisect_state()
and switch is so much nicer than having to sane_grep in BISECT_RUN.

quoted

+             strbuf_release(&command);
+             strvec_clear(&args);
+             strvec_clear(&run_args);
+
+             return res;
+     }
+}
+
 int cmd_bisect__helper(int argc, const char **argv, const char *prefix)
 {
      enum {

@@ -1086,7 +1146,8 @@ int cmd_bisect__helper(int argc, const char **argv, const char *prefix)
              BISECT_LOG,
              BISECT_REPLAY,
              BISECT_SKIP,
-             BISECT_VISUALIZE
+             BISECT_VISUALIZE,
+             BISECT_RUN,

Now this new one has the trailing comma.  I'd suggest doing so in
the previous step.

Thanks.