A list of known bugs in the MinGW port:

* cmd/src/rc
	If at the beginning of a command line is a variable definition like in
		foo=1 cat
	and the command is interrupted using ^C, then commands started
	later also see the value of foo.

	Commands like ssh.exe `eat' the console (characters typed in later
	are not visible anymore)

* lib9/mingw/fdtab
-	If a file descriptor is dup'ed, only the descriptor entry is duplicated, but
	not the file or resources. This seems to be ok for all stream-like files, but
	might cause problems for normal files, if one tried to read or write in the
	same file at different offsets via the original and a dup'ed fd. Most programs
	in p9p, though, don't seem to do this.

* cmd/src/sam:
//* cmd/du:
* lib9/mingw/dirread:
-	As there is no `inode' defined on Windows file systems,
	the qid.path member of a `Dir' struct is always set to
	zero at the moment. This can be a problem if some code
	relies on qid.path being equal for the same but not for
	different files.

	One result of qid.path being always the same can be observed
	using `du.exe -a': sometimes there are directories missing
	as there is a seen() function which checks for files already
	seen -- it is mislead by qid.path.	-- FIXED (workaround in du.c)

	In sam this situation is causing a warning about possibly
	aliased files (io.c:checkqid()), but it seems not to break
	consistency.

	Perhaps a string hash of the full path of a file could be
	used as qid.path, although there would still be a problem
	in case of collisions.

//* cmd/src/sam:
//* lib9/mingw/fdtab
//* programs using announce/listen, or pipe:
	When the file descriptor of the server side of a named pipe is
	closed, it is disconnected only, but the handle is not closed. This
	causes in an increasing number of used handles for each listen()
	or pipe() call.	-- FIXED.