* Dynamically size getcwd() buffer in get_dir function
This should fix issue 42.
* Fixup: Properly respect NO_GETCWD
* Fixup: Get rid of getcwd_error, handle NO_GETCWD in a single place
When the second argument is not a string, _file_info() wants to
ensure that there is a table on top of the stack: the second argument
or a new table. If a new table is pushed it's created on top immediately,
but if a table is passed as the second argument it can be followed
by extra arguments, with the last one ending up being used as a table,
causing a crash. The fix is to remove any potential extra arguments
using `lua_settop(L, 2)`.
Also added a few tests for this case. Ref #80.
It returns the resolved path of the symlink.
Original version by Hisham <hisham@gobolinux.org>, modified to
use a different strategy for sizing the readlink() buffer.
About indentation: currently parts of lfs.c use 2 spaces
and parts use 8. It would be nice to slowly switch to one
of styles over some time, as code is being touched here and there.
2 spaces seem preferrable as it's what Lua itself uses.
The pushresult(L, -1, ...) call will delegate to pusherror() anyway,
but this avoids the "Unused static" warning for pushresult.
Also take care of assigning something meaningful to `errno`, as
strerror(errno) is used for constructing the Lua error message.
Fixes#65
Lua 5.2 changed the protocol on how to represent closed files: Lua 5.1
sets the file pointer to NULL, Lua 5.2 sets the close function pointer
to NULL (a NULL file pointer now signals an incompletely constructed
object).
Also `luaL_checkudata` never returns NULL, it raises an error instead
if you have an invalid value (no userdata/not correct userdata type).
fs.attributes(file, 'blksize') and fs.attributes(file, 'blocks) return
the wrong values.
Compare the whole attribute name instead of the first char and remove
buggy special casing with wrong indexes into the member array.
lfs_unlock_dir can be called multiple times for the same object.
For example if lock:free is called manually.
Then lfs_unlock_dir will be called always
again, as soon as the LOCK_METATABLE is collected by GC.
This can lead to strange file errors later on, like closing
another file, which now has been assigned the same handle...
On non-win32 lfs_setmode was defined to 0, ignoring all parameters.
Now the parameters are explicitly discarded.
Fixes:
src/lfs.c: In function ‘lfs_g_setmode’:
src/lfs.c:324:7: warning: unused variable ‘op’ [-Wunused-variable]
src/lfs.c:321:47: warning: unused parameter ‘f’ [-Wunused-parameter]
To help keep the code consistent but still correct, new platform specific code needed to be introduced for the max length. On POSIX I am assuming it is that <sys/param.h> provides MAXPATHLEN. This is what the Mac/BSD man page says to use and verified this is also defined on Ubuntu Linux and Android. On Windows, MAX_PATH is used. MAX_PATH seems to still be 260 which seems kind of small; is there a different constant we are supposed to use? In both cases, the respective constants are mapped to a new #define for LFS_MAXPATHLEN to allow the code to refer to one constant and avoid any potential name collisions in case MAXPATHLEN is defined already by something else on Windows (e.g. Cygwin).
