nanodbc v3.0.0

C++ ODBC wrapper

Release notes

- The documentation's landing page is titled "nanodbc - C++ ODBC wrapper", which fits on one line where the longer wording wrapped, the heading's own anchor link taking up the width that tipped it over.
- nanodbc requires C++17. C++11 and C++14 are no longer supported, and the conditionals standing in for `std::optional`, `std::variant`, `std::any` and `std::string_view` are gone, the four being there unconditionally. The `NANODBC_HAS_STD_*` macros are still defined, so code asking whether the feature is there keeps compiling.
- `statement::bind` and `statement::bind_strings` take a `std::optional`, and a vector of them, so an absent value binds as null without a sentry value or an array of flags kept in step with the values. The sentry and flag overloads remain.
- The observers whose answer is the whole point are `[[nodiscard]]`: `result::next` and the rest of the navigation, `is_null`, `get`, `get_as` and the column accessors, along with `connected`, `rows`, `columns` and `affected_rows`. Discarding one of them was a quiet way to read the wrong row. `execute` and `transact` are left alone, ignoring their result being a fair thing to do.
- The `attribute` class that stood in for `std::variant` below C++17 is gone, as are the `#else` halves of `statement::attribute` and `connection::attribute` that went with it.
- CI builds against C++17 and C++20 and the suites run at both; the C++14 jobs are gone. The MinGW job passes the standard its matrix names, having declared one and then configured without it.
- Internals follow the standard they are now written against: fold expressions where a braced list was standing in for one, `if constexpr` where a pair of `enable_if` overloads differed only by a compile-time condition, `std::optional` read through one trait rather than two, and `using` rather than `typedef`.
- The documentation build is warning-free again: `PARAM_RETURN`'s comment gave Doxygen a code span it read as two sentences, `JAVADOC_AUTOBRIEF` ending the brief at the `?` inside it, and `doc/README.md` still described the Breathe step that generating the reference with Doxygen replaced.
- The before and after comparison on the landing page is code rather than two screenshots, so it is highlighted by the same theme as the rest of the site, selectable, searchable and legible at any width. Both halves compile, which the lint run checks: the ODBC one against the ODBC headers rather than Windows-only ones, and the nanodbc one in a narrow build and a Unicode one alike, reading its rows through a range-for rather than counting them.
- The documentation site and the API reference beside it read as one site. The pages are themed with Furo, whose colours, fonts and sizes `conf.py` sets from doxygen-awesome-css's own custom properties, and the reference follows the reader's system between light and dark the way the pages do, which being pinned light it did not.
- Describing a parameter and finding a bound column look the key up once rather than twice, `count()` followed by `at()` having asked the same map the same question either side of the answer.

Share this resource