| blob | 2def2dc02f06592da7a4b6612211312bc691a864 |
1 ;;;; stuff which is not specific to any particular build phase, but
2 ;;;; used by most of them
3 ;;;;
4 ;;;; Note: It's specifically not used when bootstrapping PCL, because
5 ;;;; we do SAVE-LISP after that, and we don't want to save extraneous
6 ;;;; bootstrapping machinery into the frozen image which will
7 ;;;; subsequently be used as the mother of all Lisp sessions.
9 ;;;; This software is part of the SBCL system. See the README file for
10 ;;;; more information.
11 ;;;;
12 ;;;; This software is derived from the CMU CL system, which was
13 ;;;; written at Carnegie Mellon University and released into the
14 ;;;; public domain. The software is in the public domain and is
15 ;;;; provided with absolutely no warranty. See the COPYING and CREDITS
16 ;;;; files for more information.
18 ;;; SB-COLD holds stuff used to build the initial SBCL core file
19 ;;; (including not only the final construction of the core file, but
20 ;;; also the preliminary steps like e.g. building the cross-compiler
21 ;;; and running the cross-compiler to produce target FASL files).
26 ;;; prefixes for filename stems when cross-compiling. These are quite arbitrary
27 ;;; (although of course they shouldn't collide with anything we don't want to
28 ;;; write over). In particular, they can be either relative path names (e.g.
29 ;;; "host-objects/" or absolute pathnames (e.g. "/tmp/sbcl-xc-host-objects/").
30 ;;;
31 ;;; The cross-compilation process will force the creation of these directories
32 ;;; by executing CL:ENSURE-DIRECTORIES-EXIST (on the xc host Common Lisp).
37 ;; Target fasl files are LOADed (actually only quasi-LOADed, in
38 ;; GENESIS) only by SBCL code, and it doesn't care about particular
39 ;; extensions, so we can use something arbitrary.
42 ;; Target fasl files from SB!C:ASSEMBLE-FILE are LOADed via GENESIS.
43 ;; The source files are compiled once as assembly files and once as
44 ;; normal lisp files. In the past, they were kept separate by
45 ;; clever symlinking in the source tree, but that became less clean
46 ;; as ports to host environments without symlinks started appearing.
47 ;; In order to keep them separate, we have the assembled versions
48 ;; with a separate suffix.
51 ;;; a function of one functional argument, which calls its functional argument
52 ;;; in an environment suitable for compiling the target. (This environment
53 ;;; includes e.g. a suitable *FEATURES* value.)
57 ;;; a function with the same calling convention as CL:COMPILE-FILE, to be
58 ;;; used to translate ordinary Lisp source files into target object files
62 ;;; designator for a function with the same calling convention as
63 ;;; SB-C:ASSEMBLE-FILE, to be used to translate assembly files into target
64 ;;; object files
66 \f
67 ;;;; some tools
69 ;;; Take the file named X and make it into a file named Y. Sorta like
70 ;;; UNIX, and unlike Common Lisp's bare RENAME-FILE, we don't allow
71 ;;; information from the original filename to influence the final
72 ;;; filename. (The reason that it's only sorta like UNIX is that in
73 ;;; UNIX "mv foo bar/" will work, but the analogous
74 ;;; (RENAME-FILE-A-LA-UNIX "foo" "bar/") should fail.)
75 ;;;
76 ;;; (This is a workaround for the weird behavior of Debian CMU CL
77 ;;; 2.4.6, where (RENAME-FILE "dir/x" "dir/y") tries to create a file
78 ;;; called "dir/dir/y". If that behavior goes away, then we should be
79 ;;; able to get rid of this function and use plain RENAME-FILE in the
80 ;;; COMPILE-STEM function above. -- WHN 19990321
84 ;; example in the ANSI spec for TRUENAME.)
87 ;; From the ANSI spec: "In this case, the file is closed
88 ;; when the truename is tried, so the truename
89 ;; information is reliable."
95 ;;; other miscellaneous tools
100 ;;; Try to minimize/conceal any non-standardness of the host Common Lisp.
102 \f
103 ;;;; special read-macros for building the cold system (and even for
104 ;;;; building some of our tools for building the cold system)
108 ;;; When cross-compiling, the *FEATURES* set for the target Lisp is
109 ;;; not in general the same as the *FEATURES* set for the host Lisp.
110 ;;; In order to refer to target features specifically, we refer to
111 ;;; *SHEBANG-FEATURES* instead of *FEATURES*, and use the #!+ and #!-
112 ;;; readmacros instead of the ordinary #+ and #- readmacros.
116 nil
128 "target features *SHEBANG-FEATURES*=~@<~S~:>~%"
142 "target backend-subfeatures *SHEBANG-BACKEND-FEATURES*=~@<~S~:>~%"
150 ;;; Some feature combinations simply don't work, and sometimes don't
151 ;;; fail until quite a ways into the build. Pick off the more obvious
152 ;;; combinations now, and provide a description of what the actual
153 ;;; failure is (not always obvious from when the build fails).
168 sb-safepoint sb-thruption sb-wtimer
169 sb-dynamic-core)))"
172 ;; Subtle memory corruption follows when sb-dynamic-core is
173 ;; active, and non-threaded allocation routines have not been
174 ;; updated to take the additional indirection into account.
175 ;; Let's avoid this unusual combination.
177 ;; There is still hope to make multithreading on DragonFly x86-64
187 failed-test-descriptions)))
188 \f
189 ;;;; cold-init-related PACKAGE and SYMBOL tools
191 ;;; Once we're done with possibly ANSIfying the COMMON-LISP package,
192 ;;; it's probably a mistake if we change it (beyond changing the
193 ;;; values of special variables such as *** and +, anyway). Set up
194 ;;; machinery to warn us when/if we change it.
195 ;;;
196 ;;; All code depending on this is itself dependent on #!+SB-SHOW.
197 #!+sb-show
201 \f
202 ;;;; master list of source files and their properties
204 ;;; flags which can be used to describe properties of source files
208 ;; cross-compiler which runs on the host ANSI Lisp. ("not host
209 ;; code", i.e. does not execute on host -- but may still be
210 ;; cross-compiled by the host, so that it executes on the target)
211 :not-host
212 ;; meaning: This file is not to be compiled as part of the target
213 ;; SBCL. ("not target code" -- but still presumably host code,
214 ;; used to support the cross-compilation process)
215 :not-target
216 ;; meaning: The #'COMPILE-STEM argument :TRACE-FILE should be T.
217 ;; When the compiler is SBCL's COMPILE-FILE or something like it,
218 ;; compiling "foo.lisp" will generate "foo.trace" which contains lots
219 ;; of exciting low-level information about representation selection,
220 ;; VOPs used by the compiler, and bits of assembly.
221 :trace-file
222 ;; meaning: This file is to be processed with the SBCL assembler,
223 ;; not COMPILE-FILE. (Note that this doesn't make sense unless
224 ;; :NOT-HOST is also set, since the SBCL assembler doesn't exist
225 ;; while the cross-compiler is being built in the host ANSI Lisp.)
226 :assem
227 ;; meaning: The #'COMPILE-STEM argument called :IGNORE-FAILURE-P
228 ;; should be true. (This is a KLUDGE: I'd like to get rid of it.
229 ;; For now, it exists so that compilation can proceed through the
230 ;; legacy warnings in src/compiler/x86/array.lisp, which I've
231 ;; never figured out but which were apparently acceptable in CMU
232 ;; CL. Eventually, it would be great to just get rid of all
233 ;; warnings and remove support for this flag. -- WHN 19990323)
245 ,@body
248 ;;; Given a STEM, remap the path component "/target/" to a suitable
249 ;;; target directory.
264 stem)))
267 ;;; Determine the source path for a stem.
272 ;;; Determine the object path for a stem/flags/mode combination.
277 (:host-compile
278 ;; On some xc hosts, it's impossible to LOAD a fasl file unless it
279 ;; has the same extension that the host uses for COMPILE-FILE
280 ;; output, so we have to be careful to use the xc host's preferred
281 ;; extension.
292 ;;; Check for stupid typos in FLAGS list keywords.
295 ;; We do duplicate stem comparison based on the object path in
296 ;; order to cover the case of stems with an :assem flag, which
297 ;; have two entries but separate object paths for each. KLUDGE:
298 ;; We have to bind *target-obj-prefix* here because it's normally
299 ;; set up later in the build process and we don't actually care
300 ;; what it is so long as it doesn't change while we're checking
301 ;; for duplicate stems.
307 ;; FIXME: We should make sure that the :assem flag is only used
308 ;; when paired with :not-host.
312 set-difference)))))
313 \f
314 ;;;; tools to compile SBCL sources to create the cross-compiler
316 ;;; a wrapper for compilation/assembly, used mostly to centralize
317 ;;; the procedure for finding full filenames from "stems"
318 ;;;
319 ;;; Compile the source file whose basic name is STEM, using some
320 ;;; standard-for-the-SBCL-build-process procedures to generate the
321 ;;; full pathnames of source file and object file. Return the pathname
322 ;;; of the object file for STEM.
323 ;;;
324 ;;; STEM and FLAGS are as per DO-STEMS-AND-FLAGS. MODE is one of
325 ;;; :HOST-COMPILE and :TARGET-COMPILE.
329 ;; Lisp Way, although it works just fine for common UNIX environments.
330 ;; Should it come to pass that the system is ported to environments
331 ;; where version numbers and so forth become an issue, it might become
332 ;; urgent to rewrite this using the fancy Common Lisp PATHNAME
333 ;; machinery instead of just using strings. In the absence of such a
334 ;; port, it might or might be a good idea to do the rewrite.
335 ;; -- WHN 19990815
351 ;; We're about to set about building a new object file. First, we
352 ;; delete any preexisting object file in order to avoid confusing
353 ;; ourselves later should we happen to bail out of compilation
354 ;; with an error.
358 ;; Original comment:
359 ;;
360 ;; Work around a bug in CLISP 1999-01-08 #'COMPILE-FILE: CLISP
361 ;; mangles relative pathnames passed as :OUTPUT-FILE arguments,
362 ;; but works OK with absolute pathnames.
363 ;;
364 ;; following discussion on cmucl-imp 2002-07
365 ;; "COMPILE-FILE-PATHNAME", it would seem safer to deal with
366 ;; absolute pathnames all the time; it is no longer clear that the
367 ;; original behaviour in CLISP was wrong or that the current
368 ;; behaviour is right; and in any case absolutifying the pathname
369 ;; insulates us against changes of behaviour. -- CSR, 2002-08-09
371 ;; (Note that this idiom is taken from the ANSI
372 ;; documentation for TRUENAME.)
375 ;; Compilation would overwrite the
376 ;; temporary object anyway and overly
377 ;; strict implementations default
378 ;; to :ERROR.
382 ;; and some compilers (e.g. OpenMCL) will complain if they're
383 ;; asked to write over a file that exists already (and isn't
384 ;; recognizeably a fasl file), so
388 ;; Try to use the compiler to generate a new temporary object file.
394 retry-compile-file
406 src)
410 obj)
412 :report report-recompile-restart
415 :report report-continue-restart
417 ;; Don't leave failed object files lying around.
421 ;; Otherwise: success, just fall through.
424 ;; If we get to here, compilation succeeded, so it's OK to rename
425 ;; the temporary output file to the permanent object file.
428 ;; nice friendly traditional return value
432 ;;; Execute function FN in an environment appropriate for compiling the
433 ;;; cross-compiler's source code in the cross-compilation host.
437 ;; the CROSS-FLOAT-INFINITY-KLUDGE, as documented in
438 ;; base-target-features.lisp-expr:
440 '(:sb-propagate-float-type
446 ;;; Process a file as source code for the cross-compiler, compiling it
447 ;;; (if necessary) in the appropriate environment, then loading it
448 ;;; into the cross-compilation host Common lisp.
456 ;;; like HOST-CLOAD-STEM, except that we don't bother to compile
460 \f
461 ;;;; tools to compile SBCL sources to create object files which will
462 ;;;; be used to create the target SBCL .core file
464 ;;; Run the cross-compiler on a file in the source directory tree to
465 ;;; produce a corresponding file in the target object directory tree.
472 ;;; (This function is not used by the build process, but is intended
473 ;;; for interactive use when experimenting with the system. It runs
474 ;;; the cross-compiler on test files with arbitrary filenames, not
475 ;;; necessarily in the source tree, e.g. in "/tmp".)