| blob | ee82c8bba2e256842a2b4a5846cd61bd1eb6dfe3 |
1 ;;;; the top level interfaces to the compiler, plus some other
2 ;;;; compiler-related stuff (e.g. CL:CALL-ARGUMENTS-LIMIT) which
3 ;;;; doesn't obviously belong anywhere else
5 ;;;; This software is part of the SBCL system. See the README file for
6 ;;;; more information.
7 ;;;;
8 ;;;; This software is derived from the CMU CL system, which was
9 ;;;; written at Carnegie Mellon University and released into the
10 ;;;; public domain. The software is in the public domain and is
11 ;;;; provided with absolutely no warranty. See the COPYING and CREDITS
12 ;;;; files for more information.
18 ;;; Set to NIL to disable loop analysis for register allocation.
21 ;;; Bind this to a stream to capture various internal debugging output.
24 ;;; The current block compilation state. These are initialized to the
25 ;;; :BLOCK-COMPILE and :ENTRY-POINTS arguments that COMPILE-FILE was
26 ;;; called with.
27 ;;;
28 ;;; *BLOCK-COMPILE-ARG* holds the original value of the :BLOCK-COMPILE
29 ;;; argument, which overrides any internal declarations.
36 ;;; When block compiling, used by PROCESS-FORM to accumulate top level
37 ;;; lambdas resulting from compiling subforms. (In reverse order.)
41 ;;; The current non-macroexpanded toplevel form as printed when
42 ;;; *compile-print* is true.
50 "When this is true, the compiler prints to *STANDARD-OUTPUT* progress
51 information about the phases of compilation of each function. (This
55 "The defaulted pathname of the file currently being compiled, or NIL if not
58 "The TRUENAME of the file currently being compiled, or NIL if not
65 ;;; the SOURCE-INFO structure for the current compilation. This is
66 ;;; null globally to indicate that we aren't currently in any
67 ;;; identifiable compilation.
70 ;;; This is true if we are within a WITH-COMPILATION-UNIT form (which
71 ;;; normally causes nested uses to be no-ops).
74 ;;; Count of the number of compilation units dynamically enclosed by
75 ;;; the current active WITH-COMPILATION-UNIT that were unwound out of.
78 ;;; Mumble conditional on *COMPILE-PROGRESS*.
97 ;; Used during compilation to map code paths to the matching
98 ;; instrumentation conses.
100 ;; Used during compilation to keep track of with source paths have been
101 ;; instrumented in which blocks.
103 ;; Stores the code coverage instrumentation results. Keys are namestrings, the
104 ;; value is a list of (CONS PATH STATE), where STATE is +CODE-COVERAGE-UNMARKED+
105 ;; for a path that has not been visited, and T for one that has.
108 \f
109 ;;;; WITH-COMPILATION-UNIT and WITH-COMPILATION-VALUES
112 "Affects compilations that take place within its dynamic extent. It is
113 intended to be eg. wrapped around the compilation of all files in the same system.
115 Following options are defined:
117 :OVERRIDE Boolean-Form
118 One of the effects of this form is to delay undefined warnings until the
119 end of the form, instead of giving them at the end of each compilation.
120 If OVERRIDE is NIL (the default), then the outermost
121 WITH-COMPILATION-UNIT form grabs the undefined warnings. Specifying
122 OVERRIDE true causes that form to grab any enclosed warnings, even if it
123 is enclosed by another WITH-COMPILATION-UNIT.
125 :POLICY Optimize-Declaration-Form
126 Provides dynamic scoping for global compiler optimization qualities and
127 restrictions, limiting effects of subsequent OPTIMIZE proclamations and
128 calls to SB-EXT:RESTRICT-COMPILER-POLICY to the dynamic scope of BODY.
130 If OVERRIDE is false, specified POLICY is merged with current global
131 policy. If OVERRIDE is true, current global policy, including any
132 restrictions, is discarded in favor of the specified POLICY.
134 Supplying POLICY NIL is equivalent to the option not being supplied at
135 all, ie. dynamic scoping of policy does not take place.
137 This option is an SBCL-specific experimental extension: Interface
138 subject to change.
140 :SOURCE-NAMESTRING Namestring-Form
141 Attaches the value returned by the Namestring-Form to the internal
142 debug-source information as the namestring of the source file. Normally
143 the namestring of the input-file for COMPILE-FILE is used: this option
144 can be used to provide source-file information for functions compiled
145 using COMPILE, or to override the input-file of COMPILE-FILE.
147 If both an outer and an inner WITH-COMPILATION-UNIT provide a
148 SOURCE-NAMESTRING, the inner one takes precedence. Unaffected
149 by :OVERRIDE.
151 This is an SBCL-specific extension.
153 :SOURCE-PLIST Plist-Form
154 Attaches the value returned by the Plist-Form to internal debug-source
155 information of functions compiled in within the dynamic extent of BODY.
157 Primarily for use by development environments, in order to eg. associate
158 function definitions with editor-buffers. Can be accessed using
159 SB-INTROSPECT:DEFINITION-SOURCE-PLIST.
161 If an outer WITH-COMPILATION-UNIT form also provide a SOURCE-PLIST, it
162 is appended to the end of the provided SOURCE-PLIST. Unaffected
163 by :OVERRIDE.
165 This is an SBCL-specific extension.
167 Examples:
169 ;; Prevent proclamations from the file leaking, and restrict
170 ;; SAFETY to 3 -- otherwise uses the current global policy.
171 (with-compilation-unit (:policy '(optimize))
172 (restrict-compiler-policy 'safety 3)
175 ;; Using default policy instead of the current global one,
176 ;; except for DEBUG 3.
177 (with-compilation-unit (:policy '(optimize debug)
178 :override t)
181 ;; Same as if :POLICY had not been specified at all: SAFETY 3
182 ;; proclamation leaks out from WITH-COMPILATION-UNIT.
183 (with-compilation-unit (:policy nil)
184 (declaim (optimize safety))
186 "
201 ;; Inside another WITH-COMPILATION-UNIT, a WITH-COMPILATION-UNIT is
202 ;; ordinarily (unless OVERRIDE) basically a no-op.
231 ;;; Is NAME something that no conforming program can rely on
232 ;;; defining?
235 (:function
238 (:type
245 ;;; This is to be called at the end of a compilation unit. It signals
246 ;;; any residual warnings about unknown stuff, then prints the total
247 ;;; error counts. ABORT-P should be true when the compilation unit was
248 ;;; aborted by throwing out. ABORT-COUNT is the number of dynamically
249 ;;; enclosed nested compilation units that were aborted.
276 (:function
278 "~@<The function ~S is undefined, and its name is ~
279 reserved by ANSI CL so that even if it were ~
280 defined later, the code doing so would not be ~
282 (:type
285 "~@<Undefined type ~S. The name starts with ~S: ~
286 probably use of a quoted type name in a context ~
287 where the name is not evaluated.~:@>"
290 "~@<Undefined type ~S. Note that name ~S is ~
291 reserved by ANSI CL, so code defining a type with ~
292 that name would not be portable.~:@>" name
293 name))))
302 "~W more use~:P of undefined ~(~A~) ~S"
303 more kind name)
305 "~W more use~:P of undefined ~(~A~) ~S"
306 more kind name))))))))))
316 abort-p)
320 "~& Undefined ~(~A~)~p:~
321 ~% ~{~<~% ~1:;~S~>~^ ~}"
324 ~[~:;~:*~& caught ~W ERROR condition~:P~]~
325 ~[~:;~:*~& caught ~W WARNING condition~:P~]~
326 ~[~:;~:*~& caught ~W STYLE-WARNING condition~:P~]~
327 ~[~:;~:*~& printed ~W note~:P~]"
336 ;; Bidrectional map between IR1/IR2/assembler abstractions
337 ;; and a corresponding small integer identifier. One direction could be done
338 ;; by adding the integer ID as an object slot, but we want both directions.
354 ;;; Evaluate BODY, then return (VALUES BODY-VALUE WARNINGS-P
355 ;;; FAILURE-P), where BODY-VALUE is the first value of the body, and
356 ;;; WARNINGS-P and FAILURE-P are as in CL:COMPILE or CL:COMPILE-FILE.
358 ;; This binding could just as well be in WITH-IR1-NAMESPACE, but
359 ;; since it's primarily a debugging tool, it's nicer to have
360 ;; a wider unique scope by ID.
376 ;;; THING is a kind of thing about which we'd like to issue a warning,
377 ;;; but showing at most one warning for a given set of <THING,FMT,ARGS>.
378 ;;; The compiler does a good job of making sure not to print repetitive
379 ;;; warnings for code that it compiles, but this solves a different problem.
380 ;;; Specifically, for a warning from PARSE-LAMBDA-LIST, there are three calls:
381 ;;; - once in the expander for defmacro itself, as it calls MAKE-MACRO-LAMBDA
382 ;;; which calls PARSE-LAMBDA-LIST. This is the toplevel form processing.
383 ;;; - again for :compile-toplevel, where the DS-BIND calls PARSE-LAMBDA-LIST.
384 ;;; If compiling in compile-toplevel, then *COMPILE-OBJECT* is a core object,
385 ;;; but if interpreting, then it is still a fasl.
386 ;;; - once for compiling to fasl. *COMPILE-OBJECT* is a fasl.
387 ;;; I'd have liked the data to be associated with the fasl, except that
388 ;;; as indicated above, the second line hides some information.
397 ;; THING is compared by EQ, FMT by STRING=.
400 ;; We don't want to walk into default values,
401 ;; e.g. (&optional (b #<insane-struct))
402 ;; because #<insane-struct> might be circular.
411 \f
412 ;;;; component compilation
415 "The upper limit on the number of times that we will consecutively do IR1
416 optimization that doesn't introduce any new code. A finite limit is
422 ;;; Repeatedly optimize COMPONENT until no further optimizations can
423 ;;; be found or we hit our iteration limit. When we hit the limit, we
424 ;;; clear the component and block REOPTIMIZE flags to discourage the
425 ;;; next optimization attempt from pounding on the same code.
469 ;;; KLUDGE: This was bumped from 5 to 10 in a DTC patch ported by MNA
470 ;;; from CMU CL into sbcl-0.6.11.44, the same one which allowed IR1
471 ;;; transforms to be delayed. Either DTC or MNA or both didn't explain
472 ;;; why, and I don't know what the rationale was. -- WHN 2001-04-28
473 ;;;
474 ;;; FIXME: It would be good to document why it's important to have a
475 ;;; large value here, and what the drawbacks of an excessively large
476 ;;; value are; and it might also be good to make it depend on
477 ;;; optimization policy.
483 ;;; Iterate doing FIND-DFO until no new dead code is discovered.
496 ;;; Do all the IR1 phases for a non-top-level component.
501 ;make this 0.
524 ;; We delay the generation of type checks until
525 ;; the type constraints have had time to
526 ;; propagate, else the compiler can confuse itself.
548 #!+immobile-code
551 ;; COMPILE-FILE puts all nontoplevel code in immobile space, but COMPILE
552 ;; offers a choice. Because the collector does not run often enough (yet),
553 ;; COMPILE usually places code in the dynamic space managed by our copying GC.
554 ;; Change this variable if your application always demands immobile code.
555 ;; The real default is set to :DYNAMIC in make-target-2-load.lisp
561 node-or-component))
581 ;; STACK only uses dominance information for DX LVAR back
582 ;; propagation (see BACK-PROPAGATE-ONE-DX-LVAR).
586 ;; Assign BLOCK-NUMBER for any cleanup blocks introduced by
587 ;; stack analysis. There shouldn't be any unreachable code after
588 ;; control, so this won't delete anything.
644 #-sb-xc-host
656 code-length
657 fixup-notes
664 code-length
665 fixup-notes
669 ;; We're done, so don't bother keeping anything around.
674 ;;; Delete components with no external entry points before we try to
675 ;;; generate code. Unreachable closures can cause IR2 conversion to
676 ;;; puke on itself, since it is the reference to the closure which
677 ;;; normally causes the components to be combined.
684 (:external
692 ;; miscellaneous sanity checks
693 ;;
694 ;; FIXME: These are basically pretty wimpy compared to the checks done
695 ;; by the old CHECK-IR1-CONSISTENCY code. It would be really nice to
696 ;; make those internal consistency checks work again and use them.
701 ;; sanity check to prevent weirdness from propagating insidiously as
702 ;; far from its root cause as it did in bug 138: Make sure that
703 ;; thing-to-COMPONENT links are consistent.
709 ;; Record xref information before optimization. This way the
710 ;; stored xref data reflects the real source as closely as
711 ;; possible.
721 #|
733 |#
735 ;; This should happen at some point before PHYSENV-ANALYZE, and
736 ;; after RECORD-COMPONENT-XREFS. Beyond that, I haven't really
737 ;; thought things through. -- AJB, 2014-Jun-08
740 ;; FIXME: What is MAYBE-MUMBLE for? Do we need it any more?
754 \f
755 ;;;; clearing global data structures
756 ;;;;
757 ;;;; FIXME: Is it possible to get rid of this stuff, getting rid of
758 ;;;; global data structures entirely when possible and consing up the
759 ;;;; others from scratch instead of clearing and reusing them?
761 ;;; Clear the INFO in constants in the *FREE-VARS*, etc. In
762 ;;; addition to allowing stuff to be reclaimed, this is required for
763 ;;; correct assignment of constant offsets, since we need to assign a
764 ;;; new offset for each component. We don't clear the FUNCTIONAL-INFO
765 ;;; slots, since they are used to keep track of functions across
766 ;;; component boundaries.
779 ;;; Blow away the REFS for all global variables, and let COMPONENT
780 ;;; be recycled.
792 x))
799 \f
800 ;;;; trace output
802 ;;; Print out some useful info about COMPONENT to STREAM.
823 \f
824 ;;; Given a pathname, return a SOURCE-INFO structure.
830 :external-format external-format
831 :subforms
836 ;;; Return a SOURCE-INFO to describe the incremental compilation of FORM.
844 ;;; Walk up the SOURCE-INFO list until we either reach a SOURCE-INFO
845 ;;; with no parent (e.g., from a REPL evaluation) or until we reach a
846 ;;; SOURCE-INFO whose FILE-INFO denotes a file.
854 finfo))))
856 ;;; If STREAM is present, return it, otherwise open a stream to the
857 ;;; current file. There must be a current file.
858 ;;;
859 ;;; FIXME: This is probably an unnecessarily roundabout way to do
860 ;;; things now that we process a single file in COMPILE-FILE (unlike
861 ;;; the old CMU CL code, which accepted multiple files). Also, the old
862 ;;; comment said
863 ;;; When we open a new file, we also reset *PACKAGE* and policy.
864 ;;; This gives the effect of rebinding around each file.
865 ;;; which doesn't seem to be true now. Check to make sure that if
866 ;;; such rebinding is necessary, it's still done somewhere.
879 :external-format external-format
880 ;; SBCL stream classes aren't available in the host
886 stream)))))
888 ;;; Close the stream in INFO if it is open.
896 ;; Loop over forms read from INFO's stream, calling FUNCTION with each.
897 ;; CONDITION-NAME is signaled if there is a reader error, and should be
898 ;; a subtype of not-so-aptly-named INPUT-ERROR-IN-COMPILE-FILE.
905 ;; Return a form read from STREAM; or for EOF use the trick,
906 ;; popularized by Kent Pitman, of returning STREAM itself.
909 ;; Reset for a new toplevel form.
917 ;; We don't need to supply :POSITION here because
918 ;; READER-ERRORs already know their position in the file.
919 :condition condition
921 ;; ANSI, in its wisdom, says that READ should return END-OF-FILE
922 ;; (and that this is not a READER-ERROR) when it encounters end of
923 ;; file in the middle of something it's trying to read,
924 ;; making it unfortunately indistinguishable from legal EOF.
925 ;; Were it not for that, it would be more elegant to just
926 ;; handle one more condition in the HANDLER-CASE.
929 condition-name
930 :condition condition
931 ;; We need to supply :POSITION here because the END-OF-FILE
932 ;; condition doesn't carry the position that the user
933 ;; probably cares about, where the failed READ began.
934 :position
937 pos)
941 stream
946 :current-index
951 current-idx))
954 ;;; Loop over FORMS retrieved from INFO. Used by COMPILE-FILE and
955 ;;; LOAD when loading from a FILE-STREAM associated with a source
956 ;;; file. ON-ERROR is the name of a condition class that should
957 ;;; be signaled if anything goes wrong during a READ.
968 ;;; Read and compile the source file.
977 ;; It's easy to get into a situation where cold-init crashes and the only
978 ;; backtrace you get from ldb is TOP-LEVEL-FORM, which means you're anywhere
979 ;; within the 23000 or so blobs of code deferred until cold-init.
980 ;; Seeing each file finish narrows things down without the noise of :sb-show,
981 ;; but this hack messes up form positions, so it's not on unless asked for.
995 ;;; Return the INDEX'th source form read from INFO and the position
996 ;;; where it was read.
1002 \f
1003 ;;;; processing of top level forms
1005 ;;; This is called by top level form processing when we are ready to
1006 ;;; actually compile something. If *BLOCK-COMPILE* is T, then we still
1007 ;;; convert the form, but delay compilation, pushing the result on
1008 ;;; *TOPLEVEL-LAMBDAS* instead.
1011 #+sb-xc-host
1015 ;; Don't bother to compile simple objects that just sit there.
1029 nil))))))
1031 ;;; Macroexpand FORM in the current environment with an error handler.
1032 ;;; We only expand one level, so that we retain all the intervening
1033 ;;; forms in the source path. A compiler-macro takes precedence over
1034 ;;; an ordinary macro as specified in CLHS 3.2.3.1
1035 ;;; Note that this function is _only_ for processing of toplevel forms.
1036 ;;; Non-toplevel forms use IR1-CONVERT-FUNCTOID which considers compiler macros.
1049 condition))))
1052 ;;; Process a PROGN-like portion of a top level form. FORMS is a list of
1053 ;;; the forms, and PATH is the source path of the FORM they came out of.
1054 ;;; COMPILE-TIME-TOO is as in ANSI "3.2.3.1 Processing of Top Level Forms".
1060 ;;; Process a top level use of LOCALLY, or anything else (e.g.
1061 ;;; MACROLET) at top level which has declarations and ordinary forms.
1062 ;;; We parse declarations and then recursively process the body.
1068 ;; FIXME: VALUES declaration
1069 ;;
1070 ;; Binding *POLICY* is pretty much of a hack, since it
1071 ;; causes LOCALLY to "capture" enclosed proclamations. It
1072 ;; is necessary because CONVERT-AND-MAYBE-COMPILE uses the
1073 ;; value of *POLICY* as the policy. The need for this hack
1074 ;; is due to the quirk that there is no way to represent in
1075 ;; a POLICY that an optimize quality came from the default.
1076 ;;
1077 ;; FIXME: Ideally, something should be done so that DECLAIM
1078 ;; inside LOCALLY works OK. Failing that, at least we could
1079 ;; issue a warning instead of silently screwing up.
1080 ;; Here's how to fix this: a POLICY object can in fact represent
1081 ;; absence of qualitities. Whenever we rebind *POLICY* (here and
1082 ;; elsewhere), it should be bound to a policy that expresses no
1083 ;; qualities. Proclamations should update SYMBOL-GLOBAL-VALUE of
1084 ;; *POLICY*, which can be seen irrespective of dynamic bindings,
1085 ;; and declarations should update the lexical policy.
1086 ;; The POLICY macro can be amended to merge the dynamic *POLICY*
1087 ;; (or whatever it came from, like a LEXENV) with the global
1088 ;; *POLICY*. COERCE-TO-POLICY can do the merge, employing a 1-line
1089 ;; cache so that repeated calls for any two fixed policy objects
1090 ;; return the identical value (since policies are immutable).
1092 ;; This is probably also a hack
1094 ;; ditto
1098 ;;; Parse an EVAL-WHEN situations list, returning three flags,
1099 ;;; (VALUES COMPILE-TOPLEVEL LOAD-TOPLEVEL EXECUTE), indicating
1100 ;;; the types of situations present in the list.
1104 '(:compile-toplevel
1105 compile
1106 :load-toplevel
1107 load
1108 :execute
1109 eval)))
1114 deprecated-names)))
1116 situations)
1121 ;;; utilities for extracting COMPONENTs of FUNCTIONALs
1132 result))))
1136 result))))
1139 &key
1140 name
1142 ;; I'd thought NIL should
1143 ;; work, but it doesn't.
1144 ;; -- WHN 2001-09-20
1155 lambda-expression
1157 ;; Convert the XEP using the policy of the real function. Otherwise
1158 ;; the wrong policy will be used for deciding whether to type-check
1159 ;; the parameters of the real function (via CONVERT-CALL /
1160 ;; PROPAGATE-TO-ARGS). -- JES, 2007-02-27
1163 :source-name source-name
1175 ;; Any leftover REFs to FUN outside local calls get replaced with the
1176 ;; XEP.
1183 xep
1184 fun)
1185 xep)))
1187 ;;; Compile LAMBDA-EXPRESSION into *COMPILE-OBJECT*, returning a
1188 ;;; description of the result.
1189 ;;; * If *COMPILE-OBJECT* is a CORE-OBJECT, then write the function
1190 ;;; into core and return the compiled FUNCTION value.
1191 ;;; * If *COMPILE-OBJECT* is a fasl file, then write the function
1192 ;;; into the fasl file and return a dump handle.
1193 ;;;
1194 ;;; If NAME is provided, then we try to use it as the name of the
1195 ;;; function for debugging/diagnostic information.
1198 &key
1199 name
1201 ;; This magical idiom seems to be the appropriate
1202 ;; path for compiling standalone LAMBDAs, judging
1203 ;; from the CMU CL code and experiment, so it's a
1204 ;; nice default for things where we don't have a
1205 ;; real source path (as in e.g. inside CL:COMPILE).
1216 :name name
1219 ;; FIXME: The compile-it code from here on is sort of a
1220 ;; twisted version of the code in COMPILE-TOPLEVEL. It'd be
1221 ;; better to find a way to share the code there; or
1222 ;; alternatively, to use this code to replace the code there.
1223 ;; (The second alternative might be pretty easy if we used
1224 ;; the :LOCALL-ONLY option to IR1-FOR-LAMBDA. Then maybe the
1225 ;; whole FUNCTIONAL-KIND=:TOPLEVEL case could go away..)
1243 result
1244 ;; KLUDGE: This code duplicates some other code in this
1245 ;; file. In the great reorganzation, the flow of program
1246 ;; logic changed from the original CMUCL model, and that
1247 ;; path (as of sbcl-0.7.5 in SUB-COMPILE-FILE) was no
1248 ;; longer followed for CORE-OBJECTS, leading to BUG
1249 ;; 156. This place is transparently not the right one for
1250 ;; this code, but I don't have a clear enough overview of
1251 ;; the compiler to know how to rearrange it all so that
1252 ;; this operation fits in nicely, and it was blocking
1253 ;; reimplementation of (DECLAIM (INLINE FOO)) (MACROLET
1254 ;; ((..)) (DEFUN FOO ...))
1255 ;;
1256 ;; FIXME: This KLUDGE doesn't solve all the problem in an
1257 ;; ideal way, as (1) definitions typed in at the REPL
1258 ;; without an INLINE declaration will give a NULL
1259 ;; FUNCTION-LAMBDA-EXPRESSION (allowable, but not ideal)
1260 ;; and (2) INLINE declarations will yield a
1261 ;; FUNCTION-LAMBDA-EXPRESSION headed by
1262 ;; SB-C:LAMBDA-WITH-LEXENV, even for null LEXENV. -- CSR,
1263 ;; 2002-07-02
1264 ;;
1265 ;; (2) is probably fairly easy to fix -- it is, after all,
1266 ;; a matter of list manipulation (or possibly of teaching
1267 ;; CL:FUNCTION about SB-C:LAMBDA-WITH-LEXENV). (1) is
1268 ;; significantly harder, as the association between
1269 ;; function object and source is a tricky one.
1270 ;;
1271 ;; FUNCTION-LAMBDA-EXPRESSION "works" (i.e. returns a
1272 ;; non-NULL list) when the function in question has been
1273 ;; compiled by (COMPILE <x> '(LAMBDA ...)); it does not
1274 ;; work when it has been compiled as part of the top-level
1275 ;; EVAL strategy of compiling everything inside (LAMBDA ()
1276 ;; ...). -- CSR, 2002-11-02
1279 #-sb-xc-host
1283 result)))
1298 form)
1307 form)
1311 ;;; Handle the evaluation the a :COMPILE-TOPLEVEL body during
1312 ;;; compilation. Normally just evaluate in the appropriate
1313 ;;; environment, but also compile if outputting a CFASL.
1328 ;; In case a null lexenv is created, it needs to get the newly
1329 ;; effective global policy, not the policy currently in *POLICY*.
1333 ;;; Process a top level FORM with the specified source PATH.
1334 ;;; * If this is a magic top level form, then do stuff.
1335 ;;; * If this is a macro, then expand it.
1336 ;;; * Otherwise, just compile it.
1337 ;;;
1338 ;;; COMPILE-TIME-TOO is as defined in ANSI
1339 ;;; "3.2.3.1 Processing of Top Level Forms".
1350 path)
1355 ;; When we're cross-compiling, consider: what should we
1356 ;; do when we hit e.g.
1357 ;; (EVAL-WHEN (:COMPILE-TOPLEVEL)
1358 ;; (DEFUN FOO (X) (+ 7 X)))?
1359 ;; DEFUN has a macro definition in the cross-compiler,
1360 ;; and a different macro definition in the target
1361 ;; compiler. The only sensible thing is to use the
1362 ;; target compiler's macro definition, since the
1363 ;; cross-compiler's macro is in general into target
1364 ;; functions which can't meaningfully be executed at
1365 ;; cross-compilation time. So make sure we do the EVAL
1366 ;; here, before we macroexpand.
1367 ;;
1368 ;; Then things get even dicier with something like
1369 ;; (DEFCONSTANT-EQX SB!XC:LAMBDA-LIST-KEYWORDS ..)
1370 ;; where we have to make sure that we don't uncross
1371 ;; the SB!XC: prefix before we do EVAL, because otherwise
1372 ;; we'd be trying to redefine the cross-compilation host's
1373 ;; constants.
1374 ;;
1375 ;; (Isn't it fun to cross-compile Common Lisp?:-)
1376 #+sb-xc-host
1381 ;; like SB!XC:DEFCONSTANT and SB!XC:DEFTYPE
1382 ;; should be equivalent to their CL: counterparts
1383 ;; when being compiled as target code. We leave
1384 ;; the rest of the form uncrossed because macros
1385 ;; might yet expand into EVAL-WHEN stuff, and
1386 ;; things inside EVAL-WHEN can't be uncrossed
1387 ;; until after we've EVALed them in the
1388 ;; cross-compilation host.)
1392 slightly-uncrossed)))
1394 ;; (Now that we're no longer processing toplevel
1395 ;; forms, and hence no longer need to worry about
1396 ;; EVAL-WHEN, we can uncross everything.)
1398 ;; (We have to demote COMPILE-TIME-TOO to NIL
1399 ;; here, no matter what it was before, since
1400 ;; otherwise we'd tend to EVAL subforms more than
1401 ;; once, because of WHEN COMPILE-TIME-TOO form
1402 ;; above.)
1404 ;; When we're not cross-compiling, we only need to
1405 ;; macroexpand once, so we can follow the 1-thru-6
1406 ;; sequence of steps in ANSI's "3.2.3.1 Processing of
1407 ;; Top Level Forms".
1408 #-sb-xc-host
1416 path
1417 compile-time-too)))))))
1419 #+sb-xc-host
1420 ;; (There are no xc EVAL-WHEN issues in the ATOM case until
1421 ;; (1) SBCL gets smart enough to handle global
1422 ;; DEFINE-SYMBOL-MACRO or SYMBOL-MACROLET and (2) SBCL
1423 ;; implementors start using symbol macros in a way which
1424 ;; interacts with SB-XC/CL distinction.)
1426 #-sb-xc-host
1432 form))))
1439 ;; CT, LT, and E here are as in Figure 3-7 of ANSI
1440 ;; "3.2.3.1 Processing of Top Level Forms".
1445 e))))
1447 body path new-compile-time-too))
1452 magic
1455 path
1456 compile-time-too
1461 magic
1464 path
1465 compile-time-too
1475 \f
1476 ;;;; load time value support
1477 ;;;;
1478 ;;;; (See EMIT-MAKE-LOAD-FORM.)
1480 ;;; Return T if we are currently producing a fasl file and hence
1481 ;;; constants need to be dumped carefully.
1486 ;;; Compile the FORMS and arrange for them to be called (for effect,
1487 ;;; not value) at load time.
1489 ;; If FORMS has exactly one PROGN containing a call of SB-PCL::SET-SLOTS,
1490 ;; then fopcompile it, otherwise use the main compiler.
1501 ;; +SLOT-UNBOUND+ is not a constant,
1502 ;; but is trivially dumpable.
1505 value-forms))
1509 ;; invoke recursive MAKE-LOAD-FORM stuff as necessary
1524 ;;; Do the actual work of COMPILE-LOAD-TIME-VALUE or
1525 ;;; COMPILE-MAKE-LOAD-FORM-INIT-FORMS.
1531 lambda)))
1533 ;;; This is called by COMPILE-TOPLEVEL when it was passed T for
1534 ;;; LOAD-TIME-VALUE-P (which happens in COMPILE-LOAD-TIME-STUFF). We
1535 ;;; don't try to combine this component with anything else and frob
1536 ;;; the name. If not in a :TOPLEVEL component, then don't bother
1537 ;;; compiling, because it was merged with a run-time component.
1546 \f
1547 ;;;; COMPILE-FILE
1557 ;;; Smash LAMBDAS into a single component, compile it, and arrange for
1558 ;;; the resulting function to be called.
1568 ;;; Compile top level code and call the top level lambdas. We pick off
1569 ;;; top level lambdas in non-top-level components here, calling
1570 ;;; SUB-c-t-l-l on each subsequence of normal top level lambdas.
1579 lambdas
1580 ;; this used to read ":start start", but
1581 ;; start can be greater than len, which
1582 ;; is an error according to ANSI - CSR,
1583 ;; 2002-04-25
1585 len)))
1594 ;;; Compile LAMBDAS (a list of CLAMBDAs for top level forms) into the
1595 ;;; object file.
1596 ;;;
1597 ;;; LOAD-TIME-VALUE-P seems to control whether it's MAKE-LOAD-FORM and
1598 ;;; COMPILE-LOAD-TIME-VALUE stuff. -- WHN 20000201
1631 ;;; Actually compile any stuff that has been queued up for block
1632 ;;; compilation.
1651 lexenv))
1652 ;; Is this right? I would think that if lexenv is null
1653 ;; we should look at *HANDLED-CONDITIONS*.
1676 ;; WOULD-MUFFLE-P is called (incorrectly) only by NOTE-UNDEFINED-REFERENCE.
1677 ;; It is not wrong per se, but as used, it is wrong, making it nearly
1678 ;; impossible to muffle a subset of undefind warnings whose NAME and KIND
1679 ;; slots match specific things tested by a user-defined predicate.
1680 ;; Attempting to do that might muffle everything, depending on how your
1681 ;; predicate responds to a vanilla WARNING. Consider e.g.
1682 ;; (AND WARNING (NOT (SATISFIES HAIRYFN)))
1683 ;; where HAIRYFN depends on the :FORMAT-CONTROL and :FORMAT-ARGUMENTS.
1691 ;;; Read all forms from INFO and compile them, with output to
1692 ;;; *COMPILE-OBJECT*. Return (VALUES ABORT-P WARNINGS-P FAILURE-P).
1703 ;; Whether to emit msan unpoisoning code depends on the runtime
1704 ;; value of the feature, not "#+msan", because we can use the target
1705 ;; compiler to compile code for itself which isn't sanitized,
1706 ;; *or* code for another image which is sanitized.
1707 ;; And we can also cross-compile assuming msan.
1737 ;; Dump the code coverage records into the fasl.
1745 code-coverage-records)
1746 list))
1747 nil
1748 nil))))
1750 nil))))
1751 ;; Some errors are sufficiently bewildering that we just fail
1752 ;; immediately, without trying to recover and compile more of
1753 ;; the input file.
1759 "~@<~@:_compilation aborted because of fatal error: ~2I~_~A~@:_~:>"
1764 ;;; Return a pathname for the named file. The file must exist.
1773 new
1789 ;;; Print some junk at the beginning and end of compilation.
1795 "preloading"
1801 file-info)
1803 :print-weekday nil
1810 won
1816 ;;; Open some files and call SUB-COMPILE-FILE. If something unwinds
1817 ;;; out of the compile, then abort the writing of the output file, so
1818 ;;; that we don't overwrite it with known garbage.
1821 &key
1823 ;; ANSI options
1825 ;; FIXME: ANSI doesn't seem to say anything about
1826 ;; *COMPILE-VERBOSE* and *COMPILE-PRINT* being rebound by this
1827 ;; function..
1832 ;; extensions
1836 "Compile INPUT-FILE, producing a corresponding fasl file and
1837 returning its filename.
1839 :PRINT
1840 If true, a message per non-macroexpanded top level form is printed
1841 to *STANDARD-OUTPUT*. Top level forms that whose subforms are
1842 processed as top level forms (eg. EVAL-WHEN, MACROLET, PROGN) receive
1843 no such message, but their subforms do.
1845 As an extension to ANSI, if :PRINT is :top-level-forms, a message
1846 per top level form after macroexpansion is printed to *STANDARD-OUTPUT*.
1847 For example, compiling an IN-PACKAGE form will result in a message about
1848 a top level SETQ in addition to the message about the IN-PACKAGE form'
1849 itself.
1851 Both forms of reporting obey the SB-EXT:*COMPILER-PRINT-VARIABLE-ALIST*.
1853 :BLOCK-COMPILE
1854 Though COMPILE-FILE accepts an additional :BLOCK-COMPILE
1855 argument, it is not currently supported. (non-standard)
1857 :TRACE-FILE
1858 If given, internal data structures are dumped to the specified
1859 file, or if a value of T is given, to a file of *.trace type
1860 derived from the input file name. (non-standard)
1862 :EMIT-CFASL
1863 (Experimental). If true, outputs the toplevel compile-time effects
1864 of this file into a separate .cfasl file."
1865 ;;; Block compilation is currently broken.
1866 #|
1867 "Also, as a workaround for vaguely-non-ANSI behavior, the
1868 :BLOCK-COMPILE argument is quasi-supported, to determine whether
1869 multiple functions are compiled together as a unit, resolving function
1870 references at compile time. NIL means that global function names are
1871 never resolved at compilation time. Currently NIL is the default
1873 of the ANSI spec allows this behavior under all circumstances, the
1874 compiler's runtime scales badly when it tries to do this for large
1875 files. If/when this performance problem is fixed, the block
1876 compilation default behavior will probably be made dependent on the
1877 SPEED and COMPILATION-SPEED optimization values, and the
1878 :BLOCK-COMPILE argument will probably become deprecated."
1879 |#
1916 default-trace-file-pathname
1918 default-trace-file-pathname))))
1952 ;; CLHS says that the first value is NIL if the "file could not
1953 ;; be created". We interpret this to mean "a valid fasl could not
1954 ;; be created" -- which can happen if the compilation is aborted
1955 ;; before the whole file has been processed, due to eg. a reader
1956 ;; error.
1958 ;; Hack around filesystem race condition...
1960 warnings-p
1961 failure-p)))
1962 \f
1963 ;;; a helper function for COMPILE-FILE-PATHNAME: the default for
1964 ;;; the OUTPUT-FILE argument
1965 ;;;
1966 ;;; ANSI: The defaults for the OUTPUT-FILE are taken from the pathname
1967 ;;; that results from merging the INPUT-FILE with the value of
1968 ;;; *DEFAULT-PATHNAME-DEFAULTS*, except that the type component should
1969 ;;; default to the appropriate implementation-defined default type for
1970 ;;; compiled files.
1974 retyped))
1976 ;;; KLUDGE: Part of the ANSI spec for this seems contradictory:
1977 ;;; If INPUT-FILE is a logical pathname and OUTPUT-FILE is unsupplied,
1978 ;;; the result is a logical pathname. If INPUT-FILE is a logical
1979 ;;; pathname, it is translated into a physical pathname as if by
1980 ;;; calling TRANSLATE-LOGICAL-PATHNAME.
1981 ;;; So I haven't really tried to make this precisely ANSI-compatible
1982 ;;; at the level of e.g. whether it returns logical pathname or a
1983 ;;; physical pathname. Patches to make it more correct are welcome.
1984 ;;; -- WHN 2000-12-09
1986 &key
1989 "Return a pathname describing what file COMPILE-FILE would write to given
1990 these arguments."
1994 \f
1995 ;;;; MAKE-LOAD-FORM stuff
1997 ;;; The entry point for MAKE-LOAD-FORM support. When IR1 conversion
1998 ;;; finds a constant structure, it invokes this to arrange for proper
1999 ;;; dumping. If it turns out that the constant has already been
2000 ;;; dumped, then we don't need to do anything.
2001 ;;;
2002 ;;; If the constant hasn't been dumped, then we check to see whether
2003 ;;; we are in the process of creating it. We detect this by
2004 ;;; maintaining the special *CONSTANTS-BEING-CREATED* as a list of all
2005 ;;; the constants we are in the process of creating. Actually, each
2006 ;;; entry is a list of the constant and any init forms that need to be
2007 ;;; processed on behalf of that constant.
2008 ;;;
2009 ;;; It's not necessarily an error for this to happen. If we are
2010 ;;; processing the init form for some object that showed up *after*
2011 ;;; the original reference to this constant, then we just need to
2012 ;;; defer the processing of that init form. To detect this, we
2013 ;;; maintain *CONSTANTS-CREATED-SINCE-LAST-INIT* as a list of the
2014 ;;; constants created since the last time we started processing an
2015 ;;; init form. If the constant passed to emit-make-load-form shows up
2016 ;;; in this list, then there is a circular chain through creation
2017 ;;; forms, which is an error.
2018 ;;;
2019 ;;; If there is some intervening init form, then we blow out of
2020 ;;; processing it by throwing to the tag PENDING-INIT. The value we
2021 ;;; throw is the entry from *CONSTANTS-BEING-CREATED*. This is so the
2022 ;;; offending init form can be tacked onto the init forms for the
2023 ;;; circular object.
2024 ;;;
2025 ;;; If the constant doesn't show up in *CONSTANTS-BEING-CREATED*, then
2026 ;;; we have to create it. We call %MAKE-LOAD-FORM and check
2027 ;;; if the result is 'FOP-STRUCT, and if so we don't do anything.
2028 ;;; The dumper will eventually get its hands on the object and use the
2029 ;;; normal structure dumping noise on it.
2030 ;;;
2031 ;;; Otherwise, we bind *CONSTANTS-BEING-CREATED* and
2032 ;;; *CONSTANTS-CREATED-SINCE- LAST-INIT* and compile the creation form
2033 ;;; much the way LOAD-TIME-VALUE does. When this finishes, we tell the
2034 ;;; dumper to use that result instead whenever it sees this constant.
2035 ;;;
2036 ;;; Now we try to compile the init form. We bind
2037 ;;; *CONSTANTS-CREATED-SINCE-LAST-INIT* to NIL and compile the init
2038 ;;; form (and any init forms that were added because of circularity
2039 ;;; detection). If this works, great. If not, we add the init forms to
2040 ;;; the init forms for the object that caused the problems and let it
2041 ;;; deal with it.
2044 ;;; FIXME: Shouldn't these^ variables be unbound outside LET forms?
2054 ;; If this is a global constant reference, we can call SYMBOL-GLOBAL-VALUE
2055 ;; during LOAD as a fasl op, and not compile a lambda.
2064 t)
2065 (:ignore-it
2066 nil)
2079 constant
2087 index))
2090 fasl)
2091 nil)
2093 constant)))
2099 collect form into forms
2101 nil)))
2105 nil)))))
2107 \f
2108 ;;;; Host compile time definitions
2109 #+sb-xc-host
2115 #+sb-xc-host