Avoid using fdropen in mz_zip_add_mem_to_archive_file_in_place or
if new flag MZ_ZIP_FLAG_READ_ALLOW_WRITING is set.
This improves performance, but also fdreopen is broken on Android
(some kind of race).
Zip archives are usually read from the back to the front, making it
possible to append them to other files (e.g. executables) in order to
provide some kind of embedded filesystem.
We can actually extract the start of the archive by looking at the
central directory offset, by comparing the specified and the actual file
offset.
The Zip64 format adds another challenge as the EOCD locator specifies
the offset to the EOCD starting from the beginning of the archive. Such
relative offset cannot be directly used for archives not starting at
position zero, hence the addition of another step that tries to locate
the 64bit EOCD right before the EOCD locator. The added overhead is
pretty small (and could be made smaller by reading both the presumed
EOCD and EOCD locator at once) and works pretty well for 90% of the
archives I've tested it with, if that heuristic fails the old behaviour
is preserved.
Miniz started including windows.h from version 3.0.0 and on when compiling for Windows using MSVC.
windows.h header file is known to conflict with C++ std namespace by defining its own min/max macros. It is a common practice to disable these ancient macros in windows.h by declaring NOMINMAX macro prior to including windows.h in code.
While this issue does not affect miniz directly due to the fact that it is straight-C code and it does not use min/max from C++ std namespace, it does affect other projects like miniz-cpp which wrap and amalgamate miniz and then both compile in C++ mode and use min/max from std namespace.
It is therefore proposed to prefix inclusion of windows.h in miniz_zip.c by the following lines:
#ifndef NOMINMAX
#define NOMINMAX
#endif
The version number in the title comment (line 1) was not updated since version 3.0.0, which causes confusion between versions 3.0.0, 3.0.1 and 3.0.2. It would be good if the title comment in this header file was updated for each release as it is exactly that this comment is being looked at by most programmers when checking the library version in their code trees.
