Re: [PATCH] push: comment on a funny unbalanced option help

[Junio C Hamano <gitster@pobox.com>]

Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes:

> On Thu, Aug 02 2018, René Scharfe wrote:
>
>> Am 02.08.2018 um 00:31 schrieb Ævar Arnfjörð Bjarmason:
>>> But looking at this again it looks like this whole thing should just be
>>> replaced by:
>>>
>>>      diff --git a/builtin/push.c b/builtin/push.c
>>>      index 9cd8e8cd56..b8fa15c101 100644
>>>      --- a/builtin/push.c
>>>      +++ b/builtin/push.c
>>>      @@ -558,9 +558,10 @@ int cmd_push(int argc, const char **argv, const char *prefix)
>>>                      OPT_BIT( 0,  "porcelain", &flags, N_("machine-readable output"), TRANSPORT_PUSH_PORCELAIN),
>>>                      OPT_BIT('f', "force", &flags, N_("force updates"), TRANSPORT_PUSH_FORCE),
>>>                      { OPTION_CALLBACK,
>>>      -                 0, CAS_OPT_NAME, &cas, N_("refname>:<expect"),
>>>      +                 0, CAS_OPT_NAME, &cas, N_("<refname>:<expect>"),
>>>                        N_("require old value of ref to be at this value"),
>>>      -                 PARSE_OPT_OPTARG, parseopt_push_cas_option },
>>>      +                 PARSE_OPT_OPTARG | PARSE_OPT_LITERAL_ARGHELP,
>>>      +                 parseopt_push_cas_option },
>>>                      { OPTION_CALLBACK, 0, "recurse-submodules", &recurse_submodules, "check|on-demand|no",
>>>                              N_("control recursive pushing of submodules"),
>>>                              PARSE_OPT_OPTARG, option_parse_recurse_submodules },
>>>
>>> I.e. the reason this is confusing is because the code originally added
>>> in 28f5d17611 ("remote.c: add command line option parser for
>>> "--force-with-lease"", 2013-07-08) didn't use PARSE_OPT_LITERAL_ARGHELP,
>>> which I also see is what read-tree etc. use already to not end up with
>>> these double <>'s, see also 29f25d493c ("parse-options: add
>>> PARSE_OPT_LITERAL_ARGHELP for complicated argh's", 2009-05-21).

Yup.  It shows that I did not know (or remember) about LIT-ARGH when
I wrote it (the line stayed in the same shape since its introduction
to the codebase), and I did not know (or remember) when I sent this
patch.  The above is the best solution to my puzzlement within the
framework of the current codebase.

>> We could check if argh comes with its own angle brackets already and
>> not add a second pair in that case, making PARSE_OPT_LITERAL_ARGHELP
>> redundant in most cases, including the one above.  Any downsides?
>> Too magical?
>
> I'm more inclined to say that we should stop using
> PARSE_OPT_LITERAL_ARGHELP in some of these cases, and change
> "refname>:<expect" to "<refname>:<expect>" in push.c, so that the help
> we emit is --force-with-lease[=<<refname>:<expect>>].

I fail to see why the outermost <> pair could be a good idea.
Without them, i.e. in what the current output shows, I can see
<refname> and <expect> are something that I should supply real
values (i.e. placeholders) and I should have a colon (literal) in
between them.  It is an established convention that a token enclosed
in a <> pair is a placeholder.

But I am not sure what you mean by <<refname>:<expect>>.

> As noted in 29f25d493c this facility wasn't added with the intent
> turning --refspec=<<refspec>> into --refspec=<refspec>, but to do stuff
> like --option=<val1>[,<val2>] for options that take comma-delimited
> options.

There is no --refspec=<<refspec>> to begin with.

A single placeholder can be written in the source as "refspec" and
shown as "--refspec=<refspec>" because you get the surrounding <>
pair for free by default.  Nobody would want to write "<refspec>" in
the arg help, as most of the option arguments are a single value
placeholder.  But if you want "<val1>[,<val2>]" in the final output,
you do *not* want surrounding <> pair, so you use the option and
write everythnig manually in the source without magic.

Or are you saying that we should consistently write surrounding "<>"
to all placeholders, and stop special casing single-token ones?
IOW, get rid of literal-arghelp and instead make that the default?

> If we're magically removing <>'s we have no consistent convention to
> tell apart --opt=<a|b|c> meaning "one of a, b or c", --refspec=<refspec>
> meaning "the literal string 'refspec'" and --refspec=<<refspec>> meaning
> add a <refspec> string, i.e. fill in your refspec here.

Ah, this is where you are off-by-one.  "--refspec=<refspec>" means
"give your refspec as its value".