Re: [PATCH v3 2/2] allow recovery from command name typos

Tay Ray Chuan [off-list ref] writes:

I have a moderately strong reaction against this; "advice" is for
helping users out of common pitfalls, and we generally do not use
the "advise" mechanism to advertise random shiny features.

I notice that with the current code, even if help.autocorrect is set
to 0 to decline the guessing, we still get "did you mean one of
these" as long as the typo is similar enough to existing command.

I am guessing that this new value `off` is a way to remedy the
situation so that users can choose to decline any guessing, and just
get "no such subcommand".  If that is the case, I think it is a vast
improvement.

+If a number is given, it will be interpreted as the deciseconds (0.1
+sec) to wait before automatically correcting and executing the mistyped
+command, with the following behaviour:
++
+* If more than one command can be deduced from the entered text, nothing
+  will be executed.

The above is from the original text, but I've always found the "can
be deduced" part hard to understand.  It is a quite roundabout way
to say we cannot guess with confidence what the user meant and avoid
committing to a wrong guess.  We may want to think a better way to
phrase the whole thing.  Perhaps something along this line:

	help.autocorrect::
		When you mistype the name of a subcommand during an
		interactive session, Git can try to guess which one
		of available subcommands you meant (Git does not
		waste cycles in a non-interactive session).  This
		configuration variable specifies what happens when
		there are one or more subcommands that you are
		likely to have meant.

        	- when set to 'ask', the choices are presented and
                  you can pick one to execute.  If the command is
                  used non-interactively,

		- when set to `off`, ...

This can be done after this patch series settles, of course.

+* If the value of this option is negative, the corrected command will be
+  executed immediately.
+* If the value is 0 - the command will be just shown but not executed.
++
+The default is to display a message suggesting that this option be set
+to `ask`, without attempting to recover (see `advice.typoPrompt`).

My comment to 'advice.typoPrompt' leads me to suggest not to change
the default to `ask`, but leave it to 0, and remove the change to
the following two files.

Nah.  No unsolicited advertisement, please.

 static int git_unknown_cmd_config(const char *var, const char *value, void *cb)
 {
-	if (!strcmp(var, "help.autocorrect"))
-		autocorrect = git_config_int(var,value);
+	if (!strcmp(var, "help.autocorrect") && value) {
+		do_advise = 0;
+		if (!strcasecmp(value, "off"))
+			;
+		else if (!strcasecmp(value, "ask"))
+			do_prompt = 1;
+		else
+			autocorrect = git_config_int(var, value);
+	}

I think the current code diagnoses

	[help]
        	autocorrect

that tries to say "true" as a syntax error.  The above simply
ignores such an entry, no?

I was hoping "off" would be usable to bypass the whole levenstein
thing, but the above code does not suggest that the remainder of
this patch would be doing that X-<.

It is the same as what is done with the current code, but if there
is no terminal available, do we even want to give this list?

+			if (isatty(2) && do_advise && advice_typo_prompt) {
+				fprintf(stderr, "\n");
+				advise(_(message_advice_prompt_ability));
+			}

Nah.  No unsolicited advertisement, please.

+		} else if (n == 1) {
+			char *in;
+			const char *ret;
+			fprintf(stderr, "\t%s\n", main_cmds.names[0]->name);
+			in = terminal_prompt(term, "[Y/n] ", 1);
+			terminal_close(term);
+			switch (in[0]) {
+			case 'y': case 'Y': case 0:
+				ret = xstrdup(main_cmds.names[0]->name);
+				clean_cmdnames(&main_cmds);
+				return ret;

OK.

+			/* otherwise, don't do anything */
+			}

Indent the comment one level deeper?

+		} else {
+			char *in;
+			const char *ret;
+			int opt;
+			for (i = 0; i < n; i++)

Can we have too many choices for this "prompt" codepath to be
practical?

+				fprintf(stderr, "%d:\t%s\n", i + 1, main_cmds.names[i]->name);
+			in = terminal_prompt(term, "[N/1/2/...] ", 1);

Would it be too much trouble to spell the actual choices out here,
instead of the ugly "/..."?

+			terminal_close(term);
+			switch (in[0]) {
+			case 'n': case 'N': case 0:
+				;
+			default:
+				opt = atoi(in);
+				if (0 < opt && opt <= n) {
+					ret = xstrdup(main_cmds.names[opt - 1]->name);
+					clean_cmdnames(&main_cmds);
+					return ret;
+				}

When the user mistypes the choice (perhaps say '8' when there are
only 7 choices available), it might be more helpful to loop here to
give him another chance.  Would such an enhancement be worth it?