Follow up be2a287f5c - it's hard to be sure from the commit message,
but I suspect the intention here was to reduce the number of indices
read in that particular situation, instead of allowing replacing
the number of real indices read direct from the table definition
with a larger number.
In any case, omitting a possible index definition which exists
is a safer approach then trying to read an index from non-index data.
Fixes#335
Use wcstombs and friends to perform Unicode conversion inside the ODBC
driver. Functionality should be identical as before, except that
installations without iconv will now be able to compile and install the
Unicode (libmdbodbcW) driver.
iconv remains present in the main libmdb library.
Defining _XOPEN_SOURCE explicitly in the source file leads to all kinds
of cross-platform headaches, plus incompatibilities between
configure-time capabilities and compile-time features. Let autoconf sort
out how to expose the largest number of functions in a way that is
consistent between configuration and compilation.
Previously we needed the cflags to include the original HAVE_ICONV value
so that MdbHandle looked the same to client applications as it did to
the compiled library. Now we use a .in file for the header so that these
definitions are not necessary when the client is compiling.
Add new mdbi_register_backend2 function with the features introduced in
PRs #321 and #322. The purpose is to prevent API/ABI breakage in the
1.0 release.
Access's `LIKE` is actually case-insensitive, but to prevent breaking existing
programs that rely on mdbtools' case-sensitive behavior, introduce a new
`ILIKE` operator to perform a case-insensitive match. Use GLib's `g_utf8_casefold`
to make the comparison UTF-8 aware. A "poor man's" version is implemented
in fakeglib, which relies on `towlower`, and won't work with multi-grapheme
case transformations (e.g. German Eszett).
Fixes#233
