how-api.texinfo

cygwin, 著名的在win32下模拟unix操作系统的东东

@section Cygwin API Questions

@subsection How does everything work?

There's a C library which provides a Unix-style API.  The
applications are linked with it and voila - they run on Windows.

The aim is to add all the goop necessary to make your apps run on
Windows into the C library.  Then your apps should run on Unix and
Windows with no changes at the source level.

The C library is in a DLL, which makes basic applications quite small.
And it allows relatively easy upgrades to the Win32/Unix translation
layer, providing that dll changes stay backward-compatible.

For a good overview of Cygwin, you may want to read the paper on Cygwin
published by the Usenix Association in conjunction with the 2d Usenix NT
Symposium in August 1998.  It is available in html format on the project
WWW site.

@subsection Are development snapshots for the Cygwin library available?

Yes.  They're made whenever anything interesting happens inside the
Cygwin library (usually roughly on a nightly basis, depending on how much
is going on).  They are only intended for those people who wish to
contribute code to the project.  If you aren't going to be happy
debugging problems in a buggy snapshot, avoid these and wait for a real
release.  The snapshots are available from
http://cygwin.com/snapshots/


@subsection How is the DOS/Unix CR/LF thing handled?

Let's start with some background.

In UNIX, a file is a file and what the file contains is whatever the
program/programmer/user told it to put into it.  In Windows, a file is
also a file and what the file contains depends not only on the
program/programmer/user but also the file processing mode.

When processing in text mode, certain values of data are treated
specially.  A \n (new line) written to the file will prepend a \r
(carriage return) so that if you `printf("Hello\n") you in fact get
"Hello\r\n".  Upon reading this combination, the \r is removed and the
number of bytes returned by the read is 1 less than was actually read.
This tends to confuse programs dependant on ftell() and fseek().  A
Ctrl-Z encountered while reading a file sets the End Of File flags even
though it truly isn't the end of file.

One of Cygwin's goals is to make it possible to easily mix Cygwin-ported
Unix programs with generic Windows programs.  As a result, Cygwin opens
files in text mode as is normal under Windows.  In the accompanying
tools, tools that deal with binaries (e.g. objdump) operate in unix
binary mode and tools that deal with text files (e.g. bash) operate in
text mode.

Some people push the notion of globally setting the default processing
mode to binary via mount point options or by setting the CYGWIN
environment variable.  But that creates a different problem.  In
binary mode, the program receives all of the data in the file, including
a \r.  Since the programs will no longer deal with these properly for
you, you would have to remove the \r from the relevant text files,
especially scripts and startup resource files.  This is a porter "cop
out", forcing the user to deal with the \r for the porter.

It is rather easy for the porter to fix the source code by supplying the
appropriate file processing mode switches to the open/fopen functions.
Treat all text files as text and treat all binary files as binary.
To be specific, you can select binary mode by adding @code{O_BINARY} to
the second argument of an @code{open} call, or @code{"b"} to second
argument of an @code{fopen} call.  You can also call @code{setmode (fd,
O_BINARY)}.

Note that because the open/fopen switches are defined by ANSI, they
exist under most flavors of Unix; open/fopen will just ignore the switch
since they have no meaning to UNIX.

Explanation adapted from mailing list email by Earnie Boyd
<earnie_boyd@@yahoo.com>.

@subsection Is the Cygwin library multi-thread-safe?

Yes.

There is also extensive support for 'POSIX threads', see the file
@code{cygwin.din} for the list of POSIX thread functions provided.

@subsection Why is some functionality only supported in Windows NT?

Windows 9x: n.
32 bit extensions and a graphical shell for a 16 bit patch to an
8 bit operating system originally coded for a 4 bit microprocessor,
written by a 2 bit company that can't stand 1 bit of competition.

But seriously, Windows 9x lacks most of the security-related calls and
has several other deficiencies with respect to its version of the Win32
API.  See the calls.texinfo document for more information as to what
is not supported in Win 9x.

@subsection How is fork() implemented?

Cygwin fork() essentially works like a non-copy on write version
of fork() (like old Unix versions used to do).  Because of this it
can be a little slow.  In most cases, you are better off using the
spawn family of calls if possible.

Here's how it works:

Parent initializes a space in the Cygwin process table for child.
Parent creates child suspended using Win32 CreateProcess call, giving
the same path it was invoked with itself.  Parent calls setjmp to save
its own context and then sets a pointer to this in the Cygwin shared
memory area (shared among all Cygwin tasks).  Parent fills in the childs
.data and .bss subsections by copying from its own address space into
the suspended child's address space.  Parent then starts the child.
Parent waits on mutex for child to get to safe point.  Child starts and
discovers if has been forked and then longjumps using the saved jump
buffer.  Child sets mutex parent is waiting on and then blocks on
another mutex waiting for parent to fill in its stack and heap.  Parent
notices child is in safe area, copies stack and heap from itself into
child, releases the mutex the child is waiting on and returns from the
fork call.  Child wakes from blocking on mutex, recreates any mmapped
areas passed to it via shared area and then returns from fork itself.

@subsection How does wildcarding (globbing) work?

If the DLL thinks it was invoked from a DOS style prompt, it runs a
`globber' over the arguments provided on the command line.  This means
that if you type @code{LS *.EXE} from DOS, it will do what you might
expect.

Beware: globbing uses @code{malloc}.  If your application defines
@code{malloc}, that will get used.  This may do horrible things to you.

@subsection How do symbolic links work?

Cygwin knows of two ways to create symlinks.

The old method is the only valid one up to but not including version 1.3.0.
If it's enabled (from 1.3.0 on by setting `nowinsymlinks' in the environment
variable CYGWIN) Cygwin generates link files with a magic header.  When you
open a file or directory that is a link to somewhere else, it opens the file
or directory listed in the magic header.  Because we don't want to have to
open every referenced file to check symlink status, Cygwin marks symlinks
with the system attribute.  Files without the system attribute are not
checked.  Because remote samba filesystems do not enable the system
attribute by default, symlinks do not work on network drives unless you
explicitly enable this attribute.

The new method which is introduced with Cygwin version 1.3.0 is enabled
by default or if `winsymlinks' is set in the environment variable CYGWIN.
Using this method, Cygwin generates symlinks by creating Windows shortcuts.
Cygwin created shortcuts have a special header (which is in that way never
created by Explorer) and the R/O attribute set.  A DOS path is stored in
the shortcut as usual and the description entry is used to store the POSIX
path.  While the POSIX path is stored as is, the DOS path has perhaps to be
rearranged to result in a valid path.  This may result in a divergence
between the DOS and the POSIX path when symlinks are moved crossing mount
points.  When a user changes the shortcut, this will be detected by Cygwin
and it will only use the DOS path then.  While Cygwin shortcuts are shown
without the ".lnk" suffix in `ls' output, non-Cygwin shortcuts are shown
with the suffix.  However, both are treated as symlinks.

Both, the old and the new symlinks can live peacefully together since Cygwin
treats both as symlinks regardless of the setting of `(no)winsymlinks' in
the environment variable CYGWIN.

@subsection Why do some files, which are not executables have the 'x' type.

When working out the unix-style attribute bits on a file, the library
has to fill out some information not provided by the WIN32 API.  

It guesses that files ending in .exe and .bat are executable, as are
ones which have a "#!" as their first characters.

@subsection How secure is Cygwin in a multi-user environment?

Cygwin is not secure in a multi-user environment.  For
example if you have a long running daemon such as "inetd"
running as admin while ordinary users are logged in, or if
you have a user logged in remotely while another user is logged
into the console, one cygwin client can trick another into
running code for it.  In this way one user may gain the
priveledge of another cygwin program running on the machine.
This is because cygwin has shared state that is accessible by 
all processes.

(Thanks to Tim Newsham (newsham@@lava.net) for this explanation).

@subsection How do the net-related functions work?

@strong{(Please note: This section has not yet been updated for the latest
net release.)}

The network support in Cygwin is supposed to provide the Unix API, not
the Winsock API.

There are differences between the semantics of functions with the same
name under the API.

E.g., the select system call on Unix can wait on a standard file handles
and handles to sockets.  The select call in winsock can only wait on
sockets.  Because of this, cygwin.dll does a lot of nasty stuff behind
the scenes, trying to persuade various winsock/win32 functions to do what
a Unix select would do.

If you are porting an application which already uses Winsock, then
using the net support in Cygwin is wrong.

But you can still use native Winsock, and use Cygwin.  The functions
which cygwin.dll exports are called 'cygwin_<name>'.  There
are a load of defines which map the standard Unix names to the names
exported by the dll -- check out include/netdb.h:

@example
..etc..
void		cygwin_setprotoent (int);
void		cygwin_setservent (int);
void		cygwin_setrpcent (int);
..etc..
#ifndef __INSIDE_CYGWIN_NET__
#define endprotoent cygwin_endprotoent 
#define endservent cygwin_endservent 
#define endrpcent  cygwin_endrpcent  
..etc..
@end example

The idea is that you'll get the Unix->Cygwin mapping if you include
the standard Unix header files.  If you use this, you won't need to
link with libwinsock.a - all the net stuff is inside the dll.

The mywinsock.h file is a standard winsock.h which has been hacked to
remove the bits which conflict with the standard Unix API, or are
defined in other headers.  E.g., in mywinsock.h, the definition of
struct hostent is removed.  This is because on a Unix box, it lives in
netdb.  It isn't a good idea to use it in your applications.

As of the b19 release, this information may be slightly out of date.

@subsection I don't want Unix sockets, how do I use normal Win32 winsock?

@strong{(Please note: This section has not yet been updated for the latest
net release.)}

To use the vanilla Win32 winsock, you just need to #define Win32_Winsock
and #include "windows.h" at the top of your source file(s).  You'll also
want to add -lwsock32 to the compiler's command line so you link against
libwsock32.a.

@subsection What version numbers are associated with Cygwin?

@strong{(Please note: This section has not yet been updated for the latest
net release.)}

There is a cygwin.dll major version number that gets incremented
every time we make a new Cygwin release available.  This
corresponds to the name of the release (e.g. beta 19's major
number is "19").  There is also a cygwin.dll minor version number.  If
we release an update of the library for an existing release, the minor
number would be incremented.

There are also Cygwin API major and minor numbers.  The major number
tracks important non-backward-compatible interface changes to the API.
An executable linked with an earlier major number will not be compatible
with the latest DLL.  The minor number tracks significant API additions
or changes that will not break older executables but may be required by
newly compiled ones.

Then there is a shared memory region compatibity version number.  It is
incremented when incompatible changes are made to the shared memory
region or to any named shared mutexes, semaphores, etc.

Finally there is a mount point registry version number which keeps track
of non-backwards-compatible changes to the registry mount table layout.
This has been "B15.0" since the beta 15 release.

@subsection Why isn't _timezone set correctly?

@strong{(Please note: This section has not yet been updated for the latest
net release.)}

Did you explicitly call tzset() before checking the value of _timezone?
If not, you must do so.

@subsection Is there a mouse interface?

There is no way to capture mouse events from Cygwin.  There are
currently no plans to add support for this.