Fix UB on empty section header [] in config file parser

[Copilot]

Parsing a config file containing [] caused undefined behavior: m_prefix was set to an empty string, then *m_prefix.rbegin() was immediately dereferenced.

Changes

• src/config_file.cpp: After extracting the section name from [...], check if m_prefix is empty and throw invalid_config_file_syntax before the rbegin() dereference:

  m_prefix = s.substr(1, s.size()-2);
  if (m_prefix.empty())
      boost::throw_exception(invalid_config_file_syntax(s, invalid_syntax::unrecognized_line));
  if (*m_prefix.rbegin() != '.')
      m_prefix += '.';

• test/parsers_test.cpp: Adds a regression test asserting that [] in a config stream throws invalid_config_file_syntax.

Original prompt

Address issue #156 in boostorg/program_options by changing the config parser so an empty section header [] is rejected as invalid syntax instead of being accepted and triggering undefined behavior.

Issue: #156

Context:

• The issue reports that boost::program_options::detail::common_config_file_iterator::get() in libs/program_options/src/config_file.cpp can form an empty m_prefix for [] and then dereference m_prefix.rbegin(), which is undefined behavior.
• The intended behavior for this fix is to reject [] as invalid config syntax.

Requested changes:

1. Update the parser implementation in libs/program_options/src/config_file.cpp so that an empty section header [] is treated as invalid syntax and does not dereference an empty string.
2. Preserve existing behavior for valid section headers.
3. Add or update automated tests to cover this case and verify that [] is rejected as invalid syntax.
4. Include any minimal necessary test fixture/config input updates.

Acceptance criteria:

• Parsing a config containing [] fails as invalid syntax.
• No undefined behavior remains for this input path.
• Existing valid section parsing behavior continues to work.
• Tests are added or updated to cover the regression.

Please create a pull request with the fix and tests.

The following is the prior conversation context from the user's chat exploration (may be truncated):

User: Can you consider issue 156?

Assistant: Yes — issue #156 looks worth considering.

My take: it appears to be a real undefined-behavior / robustness bug in common_config_file_iterator::get(), but not clearly a security issue based on the reporter’s analysis.

What the issue says

The problematic logic is:

if (*s.begin() == '[' && *s.rbegin() == ']') {
    m_prefix = s.substr(1, s.size()-2);
    if (*m_prefix.rbegin() != '.')
        m_prefix += '.';
}

For a line []:

• s.substr(1, s.size()-2) becomes an empty string
• then *m_prefix.rbegin() dereferences rbegin() on an empty string

That is a classic invalid access / UB scenario.

Why I think it should be addressed

• The report is specific and plausible.
• The reproducer is small.
• Even if it doesn’t lead to a practical exploit, it is still a parser correctness bug.
• This kind of edge case is easy to fix narrowly and safely.

Likely fix

A minimal defensive fix would be to avoid dereferencing m_prefix.rbegin() when m_prefix is empty, e.g. by checking m_prefix.empty() first.

Conceptually:

if (*s.begin() == '[' && *s.rbegin() == ']') {
    m_prefix = s.substr(1, s.size()-2);
    if (!m_prefix.empty() && *m_prefix.rbegin() != '.')
        m_prefix += '.';
}

Though semantically you may want one of two policies instead:

1. Reject [] as invalid syntax, or
2. Allow it explicitly and define what prefix it should mean.

The issue body argues the current observable behavior is not unique/useful, so rejecting [] may be the cleaner interpretation if empty section names were never intended.

Recommendation

I’d treat #156 as:

• valid bug
• low severity
• good candidate for a small fix + test

If you want, I can next:

1. inspect the relevant source file and suggest the exact patch, or
2. draft a response/comment for the issue.

User: Reject [] as invalid syntax, add a test case for that.

Assistant: I’m waiting for your confirmation on the proposed PR creation so I can continue.

User: @copilot Accepted Confirmation: Confirm cloud agent session

Assistant: Started — the PR task is now queued.

You can track it here:

data:
  - title: "Reject empty config section headers and add regression test"
    repository: boostorg/program_options
    url: https://github.com/copilot/tasks/pull/PR_kwDOAHPQcc7gWrKK?session_id=9b825e55-ecfb-477a-a043-355877cd95db
    state: queued

User: Consider issue #156

Assistant: Already in progress — I’ve started a PR task for #156 to reject [] as invalid syntax and add a regression test.

You can follow it here:

data:
  - title: "Reject empty config section headers and add regression test"
    repository: "boostorg/program_options"
    url: https://github.com/copilot/tasks/pull/PR_kwDOAHPQcc7gWrKK?session_id=9b825e55-ecfb-477a-a043-355877cd95db
    state: queued

User: That one failed with obscure internal error, can you try again?

Assistant: Yes — I retried it, and it’s now waiting on your confirmation to start a new cloud agent session.

This pull request was created from Copilot chat.