[MISC] Defer everything to parse()

[eseiler]

• Defers everything to parse()
• Refactors parse()
• Refactors init()
• No options/positional options allowed for subcommands (Follow-up)
• Fixes --export-helpXhtml being accepted for any X

---

Follow-up:

• Question: YES: Are options and flags allowed for the top-level-parser when there are subcommands?
• Question: NO: Must there always be a subcommand?

    sharg::parser parser{"./test_parser", {"-o", "true"}, /* update off */, {"foo"}};
    parser.add_option(flag_value, sharg::config{.short_id = 'o'};
    parser.parse(); // Throws because no special format + there are subcommands -> subcommand is required.

  It's currently a sharg::user_input_error (verify_subcommand()).
  I don't think it's necessary. It might be hard to differentiate between options/flags for the top-level-parser and a potentially misspelled subcommand.
• Add misspelled subcommand check back. Potentially via catching too_many_arguments from format_parse.
• subcommand is option value, e.g., ./app -o foo, where foo is a subcommand and -o is either a flag or an option
• Question: Commit 3Deferred: Do we want something like a std::monostate for the format variant?¹ To get rid of the special_format_was_set, I could also just use any non-special-format as default for the variant. But format_empty might be more expressive and help to detect errors.

¹ Can't use std::monostate directly because the visit call needs to be valid for all possible variants.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Updated (UTC)
sharg-parser	✅ Ready (Inspect)	Visit Preview	Feb 27, 2024 2:08pm

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 95.13%. Comparing base (8366490) to head (d44ad98).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #238      +/-   ##
==========================================
- Coverage   95.23%   95.13%   -0.10%     
==========================================
  Files          18       18              
  Lines        1700     1728      +28     
==========================================
+ Hits         1619     1644      +25     
- Misses         81       84       +3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[h-2]

I like it :)

[h-2]

Commit 2: Are options and flags allowed for the top-level-parser when there are subcommands?

Why not? E.g. git

[eseiler]

Commit 2: Are options and flags allowed for the top-level-parser when there are subcommands?

Why not? E.g. git

I do think it should be allowed; why shouldn't it be :)

But the code and documentation don't agree on what is allowed.

If I recall correctly from the code, there's also some very weird behavior:

Let's say I have a parser with build as subcommand, and the top-level parser has an option --name <arg>:

./app --name foo // OK
./app build // OK
./app --name foo build // throws

[h-2]

./app --name foo // OK
./app build // OK
./app --name foo build // throws

All of these should work in my opinion.

[eseiler]

The output looks like this:

../../../../test/unit/parser/format_parse_test.cpp:701: Failure
Expected equality of these values:
  expected_message
    Which is: "[...]"
  actual_message
    Which is: "[...]"
With diff:
@@ -1,3 +1,2 @@
-[...]
-[...]
+[...]

Google Test trace:
../../../../test/unit/parser/format_parse_test.cpp:719: check_error

Without the trace, it would report line 701, i.e., the EXPECT_EQ in the lambda.
Not very helpful.

The trace adds the last part, which then correctly points to line 719.

parser = /* ... */;
{
    SCOPED_TRACE("test1");
    check_error();
}

parser = /* ... */;
{
    SCOPED_TRACE("test2");
    check_error();
}

Would also work instead, but too many scopes for my test.

We might unify the check_error thing at some point since we use it in multiple locations.

[smehringer]

sorry, whats wrong here now? I remember you said it must be deleted but I forget and can't see why

[eseiler]

Originally because options are no longer allowed.
However, I now changed it to just use a flag; and also changed The following options below to The following options