Switch branches/tags
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
681 lines (530 sloc) 23.5 KB
;;;; -*- Lisp -*-
;;;; This software is part of the SBCL system. See the README file for
;;;; more information.
;;;; This software is derived from the CMU CL system, which was
;;;; written at Carnegie Mellon University and released into the
;;;; public domain. The software is in the public domain and is
;;;; provided with absolutely no warranty. See the COPYING and CREDITS
;;;; files for more information.
;;; a linear ordering of system sources which works both to
;;; compile/load the cross-compiler under the host Common Lisp and
;;; then to cross-compile the complete system into the
;;; under-construction target SBCL
;;; The keyword flags (:NOT-HOST, :NOT-TARGET, :ASSEM...) are
;;; documented in the code which implements their effects. (As of
;;; sbcl-0.7.10, the comments are on DEFPARAMETER *EXPECTED-STEM-FLAGS*
;;; in src/cold/shared.lisp.)
;;; Of course, it'd be very nice to have this be a dependency DAG
;;; instead, so that we could do automated incremental recompilation.
;;; But the dependencies are varied and subtle, and it'd be extremely
;;; difficult to extract them automatically, and it'd be extremely
;;; tedious and error-prone to extract them manually, so we don't
;;; extract them. (It would be nice to fix this someday. The most
;;; feasible approach that I can think of would be to make the
;;; dependencies work on a package level, not an individual file
;;; level. Doing it at the package level would make the granularity
;;; coarse enough that it would probably be pretty easy to maintain
;;; the dependency information manually, and the brittleness of the
;;; package system would help make most violations of the declared
;;; dependencies obvious at build time. -- WHN 20000803
;;; miscellaneous
;; This comes early because it's useful for debugging everywhere.
;; This comes early because the cross-compilation host's backquote
;; logic can expand into something which can't be executed on the
;; target Lisp (e.g. in CMU CL where it expands into internal
;; functions like BACKQ-LIST), and by replacing the host backquote
;; logic with our own as early as possible, we minimize the chance of
;; any forms referring to cross-compilation host internal functions
;; leaking into target SBCL code.
;; It's difficult to be too early with a DECLAIM SPECIAL (or DEFVAR
;; or whatever) thanks to the sullenly-do-the-wrong-thing semantics
;; of CL special binding when the variable is undeclared.
("src/code/globals" :not-host)
;; various DEFSETFs and/or other DEFMACROish things, defined as early as
;; possible so we don't need to fiddle with any subtleties of defining them
;; before any possible use
;; KLUDGE: It would be nice to reimplement most or all of these as
;; functions (possibly inlined functions) so that we wouldn't need to
;; worry so much about forcing them all to be defined before any possible
;; use. It might be pretty tedious, though, working through any
;; transforms and translators and optimizers and so forth to make sure
;; that they can handle the change. -- WHN 19990919
;;; cross-compiler-only replacements for stuff which in target Lisp would be
;;; supplied by basic machinery
("src/code/cross-misc" :not-target)
("src/code/cross-byte" :not-target)
("src/code/cross-float" :not-target)
("src/code/cross-io" :not-target)
("src/code/cross-sap" :not-target)
("src/code/cross-make-load-form" :not-target)
;;; stuff needed early both in cross-compilation host and in target Lisp
;; comes early so that stuff can reason about function names
;; for various constants e.g. SB!XC:MOST-POSITIVE-FIXNUM and
;; SB!VM:N-LOWTAG-BITS, needed by "early-objdef" and others
("src/code/early-array") ; needs "early-vm" numbers
("src/code/early-extensions") ; on host for COLLECT, SYMBOLICATE, etc.
("src/code/parse-body") ; on host for PARSE-BODY
("src/code/parse-defmacro") ; on host for PARSE-DEFMACRO
("src/compiler/deftype") ; on host for SB!XC:DEFTYPE
("src/code/early-alieneval") ; for vars needed both at build and run time
;; mostly needed by stuff from comcom, but also used by "x86-vm"
;;; basic machinery for the target Lisp. Note that although most of these
;;; files are flagged :NOT-HOST, a few might not be.
("src/code/target-defbangmethod" :not-host)
("src/code/early-print" :not-host)
("src/code/early-pprint" :not-host)
("src/code/early-impl" :not-host)
("src/code/target-extensions" :not-host)
("src/code/early-defstructs" :not-host) ; gotta-be-first DEFSTRUCTs
("src/code/funutils" :not-host)
;; This needs DEF!STRUCT, and is itself needed early so that structure
;; accessors and inline functions defined here can be compiled inline
;; later. (Avoiding full calls not only increases efficiency, but also
;; avoids some cold init issues involving full calls to structure
;; accessors.)
("src/code/pcounter" :not-host)
("src/code/ansi-stream" :not-host)
("src/code/sysmacs" :not-host)
;; "assembly/assemfile" was here in the sequence inherited from
;; CMU CL worldcom.lisp, but also appears later in the sequence
;; inherited from CMU CL comcom.lisp. We shouldn't need two versions,
;; so I've deleted the one here. -- WHN 19990620
("src/code/target-error" :not-host)
;; a comment from classic CMU CL:
;; "These guys can supposedly come in any order, but not really.
;; Some are put at the end so that macros don't run interpreted
;; and stuff."
;; Dunno exactly what this meant or whether it still holds. -- WHN 19990803
;; FIXME: more informative and up-to-date comment?
("src/code/kernel" :not-host)
("src/code/toplevel" :not-host)
("src/code/cold-error" :not-host)
("src/code/fdefinition" :not-host)
;; FIXME: Figure out some way to make the compiler macro for INFO
;; available for compilation of "code/fdefinition".
;; In classic CMU CL, code/type was here. I've since split that into
;; lots of smaller pieces, some of which are here and some of which
;; are handled later in the sequence, when the cross-compiler is
;; built. -- WHN 19990620
("src/code/target-type" :not-host)
("src/code/pred" :not-host)
("src/code/target-alieneval" :not-host)
("src/code/target-c-call" :not-host)
("src/code/target-allocate" :not-host)
;; This needs DEFINE-ALIEN-ROUTINE from target-alieneval.
("src/code/misc-aliens" :not-host)
("src/code/array" :not-host)
("src/code/target-sxhash" :not-host)
("src/code/list" :not-host)
("src/code/seq" :not-host) ; "code/seq" should come after "code/list".
("src/code/coerce" :not-host)
("src/code/string" :not-host)
("src/code/mipsstrops" :not-host)
;; "src/code/unix.lisp" needs this. It's generated automatically by
;; grovel_headers.c, i.e. it's not in CVS.
("output/stuff-groveled-from-headers" :not-host)
("src/code/unix" :not-host)
#!+mach ("src/code/mach" :not-host)
#!+mach ("src/code/mach-os" :not-host)
#!+sunos ("src/code/sunos-os" :not-host)
#!+hpux ("src/code/hpux-os" :not-host)
#!+osf1 ("src/code/osf1-os" :not-host)
#!+irix ("src/code/irix-os" :not-host)
#!+bsd ("src/code/bsd-os" :not-host)
#!+linux ("src/code/linux-os" :not-host)
;; sparc-vm and ppc-vm need sc-offset defined to get at internal
;; error args. This file contains stuff previously in
;; debug-info.lisp. Should it therefore be :not-host? -- CSR,
;; 2002-02-05
;; KLUDGE: I'd prefer to have this done with a "code/target" softlink
;; instead of a bunch of reader macros. -- WHN 19990308
#!+sparc ("src/code/sparc-vm" :not-host)
#!+hppa ("src/code/hppa-vm" :not-host)
#!+x86 ("src/code/x86-vm" :not-host)
#!+ppc ("src/code/ppc-vm" :not-host)
#!+alpha ("src/code/alpha-vm" :not-host)
#!+mips ("src/code/mips-vm" :not-host)
;; FIXME: do we really want to keep this? -- CSR, 2002-08-31
#!+rt ("src/code/rt-vm" :not-host)
("src/code/target-signal" :not-host) ; needs OS-CONTEXT-T from x86-vm
("src/code/symbol" :not-host)
("src/code/bignum" :not-host)
("src/code/numbers" :not-host)
("src/code/float-trap" :not-host)
("src/code/float" :not-host)
("src/code/irrat" :not-host)
("src/code/target-char" :not-host)
("src/code/target-misc" :not-host)
("src/code/room" :not-host)
("src/code/stream" :not-host)
("src/code/print" :not-host)
("src/code/pprint" :not-host)
("src/code/target-format" :not-host)
("src/code/defpackage" :not-host)
("src/code/pp-backq" :not-host)
("src/code/error-error" :not-host) ; needs WITH-STANDARD-IO-SYNTAX macro
("src/code/serve-event" :not-host)
("src/code/fd-stream" :not-host)
("src/code/module" :not-host)
("src/code/interr" :not-host)
("src/code/query" :not-host)
("src/code/sort" :not-host)
("src/code/time" :not-host)
("src/code/weak" :not-host)
("src/code/final" :not-host)
("src/code/setf-funs" :not-host)
("src/code/stubs" :not-host)
("src/code/exhaust" :not-host)
;;; compiler (and a few miscellaneous files whose dependencies make it
;;; convenient to stick them here)
;; ("src/code/defbangmacro" was here until sbcl-
;; needed by "compiler/vop"
;; for e.g. BLOCK-ANNOTATION, needed by "compiler/vop"
;; for e.g. PRIMITIVE-TYPE, needed by "vmdef"
;; needed by "vm" and "primtype"
;; for e.g. MAX-VOP-TN-REFS, needed by "meta-vmdef"
;; needs "backend"
;; for INFO and SB!XC:MACRO-FUNCTION, needed by defmacro.lisp
;; for e.g. !DEF-PRIMITIVE-TYPE, needed by primtype.lisp, and
;; DEFINE-STORAGE-CLASS, needed by target/vm.lisp
;; for e.g. DESCRIPTOR-REG, needed by primtype.lisp
;; for e.g. SPECIFIER-TYPE, needed by primtype.lisp
;; FIXME: Classic CMU CL had (OPTIMIZE (SAFETY 2) (DEBUG 2) declared
;; around the compilation of "code/class". Why?
;; The definition of CONDITION-CLASS depends on SLOT-CLASS, defined
;; in class.lisp.
("src/code/condition" :not-host)
;; the implementation of the compiler-affecting part of forms like
;; DEFMACRO and DEFTYPE; must be loaded before we can start
;; defining types
;; The following two files trigger function/macro redefinition
;; warnings in clisp during make-host-2; as a workaround, we ignore
;; the failure values from COMPILE-FILE under clisp.
;; for DEFSTRUCT ALIEN-TYPE, needed by host-type.lisp
("src/code/host-alieneval" #+clisp :ignore-failure-p)
;; can't be done until definition of e.g. DEFINE-ALIEN-TYPE-CLASS in
;; host-alieneval.lisp
("src/code/host-c-call" #+clisp :ignore-failure-p)
;; SB!XC:DEFTYPE is needed in order to compile late-type
;; in the host Common Lisp, and in order to run, it needs
;; These appear here in the build sequence because they require
;; * the macro INFO, defined in globaldb.lisp, and
;; * the function PARSE-DEFMACRO, defined in parse-defmacro.lisp,
;; and because they define
;; * the function SPECIFIER-TYPE, which is used in fndb.lisp.
;; defines IR1-ATTRIBUTES macro, needed by proclaim.lisp
;; needs FUN-INFO structure slot setters, defined in knownfun.lisp
;; stuff needed by "code/defstruct"
("src/code/cross-type" :not-target)
;; The DEFSTRUCT machinery needs SB!XC:SUBTYPEP, defined in
;; "code/late-type", and SB!XC:TYPEP, defined in "code/cross-type",
;; and SPECIALIZE-ARRAY-TYPE, defined in "compiler/generic/vm-type",
;; and SB!XC:PROCLAIM, defined in "src/compiler/proclaim"
("src/code/target-defstruct" :not-host)
;; ALIEN-VALUE has to be defined as a class (done by DEFSTRUCT
;; machinery) before we can set its superclasses here.
;; This needs not just the SB!XC:DEFSTRUCT machinery, but also
;; the TYPE= stuff defined in late-type.lisp, and the
;; CHECK-FUN-NAME defined in proclaim.lisp.
("src/code/typep" :not-host)
;; These define target types needed by fndb.lisp.
;; KLUDGE: Much stuff above here is the type system and/or the INFO
;; system, not really the compiler proper. It might be easier to
;; understand the system if those things were split off into packages
;; SB-TYPE and SB-INFO and built in their own sections. -- WHN 20000124
;; In classic CMU CL (re)build order, these were done later, but
;; in building from scratch, these must be loaded before
;; "compiler/generic/objdef" in order to allow forms like
;; (DEFINE-PRIMITIVE-OBJECT (..) (CAR ..) ..) to work.
;; This has ASSEMBLY-UNIT-related stuff needed by core.lisp.
;; core.lisp contains DEFSTRUCT CORE-OBJECT, and "compiler/main.lisp"
;; does lots of (TYPEP FOO 'CORE-OBJECT), so it's nice to compile this
;; before "compiler/main.lisp" so that those can be coded efficiently
;; (and so that they don't cause lots of annoying compiler warnings
;; about undefined types).
("src/code/cross-thread" :not-target)
("src/code/fop") ; needs macros from code/load.lisp
("src/compiler/trace-table") ; needs EMIT-LABEL macro from compiler/assem.lisp
;; Compiling this requires fop definitions from code/fop.lisp and
;; trace table definitions from compiler/trace-table.lisp.
("src/compiler/main") ; needs DEFSTRUCT FASL-OUTPUT from dump.lisp
("src/compiler/target-main" :not-host)
;; Compiling this file requires the macros SB!ASSEM:EMIT-LABEL and
;; SB!ASSEM:EMIT-POST-IT, defined in assem.lisp, and also possibly
;; the definition of the LOCATION-INFO structure (if structures in
;; the host lisp have setf expanders rather than setf functions).
("src/compiler/fixup") ; for DEFSTRUCT FIXUP, used by insts.lisp
;; KLUDGE: for ppc and sparc this appears to be necessary -- see the
;; comment below regarding src/compiler/target/array -- CSR,
;; 2002-05-05
;; src/compiler/sparc/c-call contains a deftransform for
;; KLUDGE: Compiling this file for X86 raises alarming warnings of
;; the form
;; Argument FOO to VOP CHECK-BOUND has SC restriction
;; DESCRIPTOR-REG which is not allowed by the operand type:
;; This seems not to be something that I broke, but rather a "feature"
;; inherited from classic CMU CL. (Debian cmucl_2.4.8.deb compiling
;; Debian cmucl_2.4.8.tar.gz raises the same warning). Thus, even though
;; these warnings are severe enough that they would ordinarily abort
;; compilation, for now we blithely ignore them and press on to more
;; pressing problems. Someday, though, it would be nice to figure out
;; what the problem is and fix it. (See the comments in
;; src/compiler/x86/array for a candidate patch.) -- WHN 19990323
;; KLUDGE: The assembly files need to be compiled twice: once as
;; normal lisp files, and once by sb-c:assemble-file. We play some
;; symlink games to make sure we don't scribble over anything we
;; shouldn't, but these are actually the same files:
("src/assembly/target/assem-rtns" :assem :not-host)
("src/assembly/target/array" :assem :not-host)
("src/assembly/target/arith" :assem :not-host)
("src/assembly/target/alloc" :assem :not-host)
#!+sb-dyncount ("src/compiler/dyncount")
#!+sb-dyncount ("src/code/dyncount")
;; needed by OPEN-FASL-OUTPUT, which is called by COMPILE-FILE
;; needed by various unhappy-path cases in the cross-compiler
;; This wasn't in classic CMU CL "comcom.lisp", but it has some stuff
;; that Python-as-cross-compiler has turned out to need.
;; files which depend in some way (directly or indirectly) on stuff
;; compiled as part of the compiler
("src/code/late-extensions") ; needs condition system
("src/compiler/generic/target-core" :not-host) ; uses stuff from
; "compiler/generic/core"
("src/code/eval" :not-host) ; uses INFO, wants compiler macro
("src/code/target-sap" :not-host) ; uses SAP-INT type
("src/code/target-package" :not-host) ; needs "code/package"
("src/code/target-random" :not-host) ; needs "code/random"
("src/code/target-hash-table" :not-host) ; needs "code/hash-table"
("src/code/reader" :not-host) ; needs "code/readtable"
("src/code/target-pathname" :not-host) ; needs "code/pathname"
("src/code/filesys" :not-host) ; needs HOST from "code/pathname"
("src/code/save" :not-host) ; uses the definition of PATHNAME
; from "code/pathname"
("src/code/sharpm" :not-host) ; uses stuff from "code/reader"
("src/code/target-thread" :not-host)
("src/code/target-unithread" :not-host)
;; defines SB!DI:DO-DEBUG-FUN-BLOCKS, needed by target-disassem.lisp
("src/code/gc" :not-host)
("src/code/purify" :not-host)
("src/code/debug-int" :not-host)
;; target-only assemblerish stuff
("src/compiler/target-disassem" :not-host)
("src/compiler/target/target-insts" :not-host)
("src/code/debug" :not-host)
;; The code here can't be compiled until CONDITION and
;; declared special.
("src/code/bit-bash" :not-host) ; needs %NEGATE from assembly/target/arith
("src/code/target-load" :not-host) ; needs special vars from code/load.lisp
("src/compiler/target/sanctify" :not-host)
;; FIXME: Does this really need stuff from compiler/dump.lisp?
("src/compiler/target-dump" :not-host) ; needs stuff from compiler/dump.lisp
("src/code/cold-init" :not-host) ; needs (SETF EXTERN-ALIEN) macroexpansion
;; target macros and DECLAIMs installed at build-the-cross-compiler time
;; Declare all target special variables defined by ANSI now, so that
;; we don't have to worry about any of them being bound incorrectly
;; when the compiler processes code which appears before the appropriate
;; FIXME: here? earlier? can probably be as late as possible. Also
;; fundamental target macros (e.g. CL:DO and CL:DEFUN) and support
;; for them
;; other target-code-building stuff which can't be processed until
;; machinery like SB!XC:DEFMACRO exists
("src/code/late-format") ; needs SB!XC:DEFMACRO
("src/code/sxhash") ; needs SB!XC:DEFINE-MODIFY-MACRO
;; PCL-related stuff, which shouldn't need to be done earlier than
;; anything else in cold build because after all it used to be
;; postponed 'til warm init with no problems.