Remove duplicate declaration

[sbcl.git] / contrib / sb-posix / README

-*- Text -*-

Note: this file's description of a POSIX binding for Common Lisp does
not necessarily describe the state of the sb-posix module.  If you're
looking for a description sb-posix, look at sb-posix.texinfo in this
directory.

* Scope

The scope of this interface is "operating system calls on a typical
Unixlike platform".  This is section 2 of the Unix manual, plus
section 3 calls that are (a) typically found in libc, but (b) not part
of the C standard.  For example, we intend to provide support for
opendir() and readdir(), but not for printf().  Note that we do not
assert or imply any claim of conformance with the official standards
issued by POSIX groups or SVID or anyone else: we're simply using
"POSIX" in a generic sense to refer to Unix and Unix-like systems.

Some facilities are omitted where they offer absolutely no additional
use over some portable function, or would be actively dangerous to the
consistency of Lisp.  Not all functions are available on all
platforms.  [TBD: unavailable functions should (a) not exist, or (b)
exist but signal some kind of "not available on this platform" error]

The general intent is for a low-level interface.  There are three
reasons for this: it's easier to write a high-level interface given a
low-level one than vice versa, there are fewer philosophical
disagreements about what it should look like, and the work of
implementing it is easily parallelisable - and in fact, can be
attempted on an as-needed basis by the various people who want it.

* Symbol names

In SBCL this interface is in the SB-POSIX package.  This package
contains a Lisp function for each supported Unix function, and a
variable or constant for each supported unix constant.  A symbol name
is derived from the C binding's name, by (a) uppercasing, then (b)
removing leading underscores (#\_) then replacing remaining underscore
characters with the hyphen (#\-). The requirement to uppercase is so
that in a standard upcasing reader the user may write posix:creat
instead of posix:|creat| as would otherise be required - some
extra-standard implementations may have alternative means to achieve
the same effect.

No other changes to "Lispify" symbol names are made, so creat()
becomes CREAT, not CREATE

The user is encouraged not to (USE-PACKAGE :SB-POSIX) but instead to
use the SB-POSIX: prefix on all references, as some of our symbols
have the same name as CL symbols (OPEN, CLOSE, SIGNAL etc).

[ Rationale: We use similar names to the C bindings so that unix
manual pages can be used for documentation.  To avoid name clashes
with CL or other functions, the approaches considered were (a) prefix
the names of clashing symbols with "POSIX-" or similar, (b) prefix
_all_ symbols with "POSIX-", (c) use the packaging system to resolve
ambiguities.  (a) was rejected as the set of symbols we may
potentially clash with is not fixed (for example, if new symbols are
added to SB-EXT) so symbols might have to be renamed over the lifetime
of SB-POSIX, which is not good.  The choice between (b) and (c) was
made on the grounds that POSIX-OPEN is about as much typing as
SB-POSIX:OPEN anyway, and symbol munging is, the author feels,
slightly tacky, when there's a package system available to do it more
cleanly ]


* Types

Some functions accept objects such as filenames or file
descriptors. In the C binding to POSIX these are represented as
strings and small integers respectively. For the Lisp programmer's
convenience we introduce designators such that CL pathnames or open
streams can be passed to these functions.

** file-descriptor

A file-descriptor is a non-negative small integer.  

A file-stream is a designator for a file-descriptor: the streams file
descriptor is extracted.  Note that mixing io operations on a stream 
with operations directly on its descriptor may produce unexpected
results if the stream is buffered.

** filename

A filename is a string.  

A pathname is a designator for a filename: the filename is computed
using the same mechanism as the implementation would use to map
pathnames to OS filenames internally.

In an implementation that supports pathnames to files on other hosts, 
using mechanisms not available to the underlying OS (for example, 
using an FTP or HTTP client in the Lisp implementation), the effect 
of supplying this interface with a pathname to such a file is undefined.


** buffer

A buffer is an opaque object which represents an area of memory that
system calls may access.  It has accessors BUFFER-START and
BUFFER-LENGTH, and can be created using ALLOCATE-BUFFER or GET-BUFFER.

[ TODO: GET-BUFFER is a silly name.  Come up with a better one ]

The object NIL is a designator for a buffer, meaning a NULL pointer.

A vector of (UNSIGNED-BYTE 8) is a designator for a buffer: it is
converted to a buffer of appropriate start and length using an
identity mapping.  This may or may not involve creating a copy of the
data.

A vector of CHARACTER is a designator for a buffer: it is converted to
a buffer of appropriate start and length using an implementation-
defined transformation that obviously depends on the implementation's
representation of characters.  This may or may not involve creating a
copy of the data.  

Implementations may optionally extend these designators to include
other types - for example, types that already exist in the
implementation's FFI.

** Structures, unions

C structures and unions are opaque to Lisp and may be implemented
using any appropriate method. Structure names are chosen according to
the general rules for symbols.

Accessors must be provided for each documented field in the
structure. These are named structure-name-field-name where the two
components are chosen according to the general rules, with the
exception that in cases where all fields in a given structure have a
common prefix, that prefix is omitted. For example, stat.st_dev in C
becomes STAT-DEV in Lisp.

For any structure that the user may need to allocate himself, there
must also be a MAKE-structure-name function. This takes keyword
arguments with names deriving from each documented field name
according to the general rules for symbols.

[ TDB: GC issues for buffers/structures/unions: probably a
WITHOUT-MOVING macro or similar that will stop them from being moved
or collected during its extent ]


** Type conversion functions

For each of these types there is a function of the same name that
converts any valid designator for the type into an object of said type.

This example is merely an example: actual output will vary between
systems, implementations and environments

(with-open-file (o "/tmp/foo" :direction :output) 
  (sb-posix:file-descriptor o)) 
=> 4

[ TBD: some memorable and nicely uniform protocol for transforming
objects of these new types into instances of the Lisp-friendly types
that may designate them: e.g how to get a stream from a fd ]


* Function parameters

The calling convention is modelled after that of CMUCL's UNIX package:
in particular, it's like the C interface except that

a) length arguments are omitted or optional where the sensible value
is obvious.  For example, 

(read fd buffer &optional (length (length buffer))) => bytes-read

b) where C simulates "out" parameters using pointers (for instance, in
pipe() or socketpair()) these may be optional or omitted in the Lisp
interface: if not provided, appropriate objects will be allocated and
returned (using multiple return values if necessary).

c) some functions accept objects such as filenames or file
descriptors.  Wherever these are specified as such in the C bindings,
the Lisp interface accepts designators for them as specified in the
'Types' section above

[ Rationale: Keeping exact 1:1 correspondence with C conventions is
less important here, as the function argument list can easily be
accessed to find out exactly what the arguments are.  Designators
are primarily a convenience feature ]

* Function return values

The return value is usually the same as for the C binding, except in
error cases: where the C function is defined as returning some
sentinel value and setting "errno" on error, we instead signal an
error of type SYSCALL-ERROR.  The actual error value ("errno") is
stored in this condition and can be accessed with SYSCALL-ERRNO.
[TBA: some interface to strerror, to get the user-readable translation
of the error number]

We do not automatically translate the returned value into "Lispy"
objects - for example, SB-POSIX:OPEN returns a small integer, not a
stream.  Exception: boolean-returning functions (or, more commonly,
macros) do not return a C integer, but instead a lisp boolean [ or
maybe "true"/"false" in CLtS parlance ]; the rationale behind this
exception is that there is nothing that can be meaningfully done with
the boolean except test for truth or falsity -- it cannot be passed
meaningfully to other POSIX functions.

[ Rationale: This is an interface to POSIX, not a high-level interface
that uses POSIX, and many people using it will actually want to mess
with the file descriptors directly.  People needing Lispy interfaces
can implement them atop this - or indeed, use the existing COMMON-LISP
package, which already has many high-level constructs built on top of
the operating system ;-) ]


* Implementation

The initial implementation is in contrib/sb-posix, and being filled
out on an as-needed basis.  Contributions following these style rules
are welcome from anyone who writes them, provided the author is happy
to release the code as Public Domain or MIT-style licence.

See/update the TODO list for current status

** Designators

See designator.lisp, add a define-designator form

** Adding functions

The use of DEFINE-CALL macro in interface.lisp should be obvious from
the existing examples, if less so from the macroexpansion





GC issues

buffers that refer to C stuff are probably not movable by GC anyway

a buffer that refers to a Lisp object may have trouble if the Lisp
object is moved