X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;f=src%2Fcold%2Fansify.lisp;h=901286d9f9570423a884035b36b176e3219b7092;hb=f0f3805c145f2699701997761e2c6f55c475c192;hp=8e05fd49028a4cfd9e50c9a7e1d8104553db7c41;hpb=a530bbe337109d898d5b4a001fc8f1afa3b5dc39;p=sbcl.git diff --git a/src/cold/ansify.lisp b/src/cold/ansify.lisp index 8e05fd4..901286d 100644 --- a/src/cold/ansify.lisp +++ b/src/cold/ansify.lisp @@ -1,4 +1,4 @@ -;;;; patches to hide some implementation idiosyncrasies in our +;;;; patches to work around implementation idiosyncrasies in our ;;;; cross-compilation host ;;;; This software is part of the SBCL system. See the README file for @@ -9,43 +9,44 @@ ;;;; 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. + +;;;; CLISP issues -;;;; $Header$ - +;;; as explained on #lisp ca. October 2003: +;;; chandler: nope, I'm blaming another clisp bug +;;; [8]> least-positive-short-float +;;; 2.93874s-39 +;;; [9]> (coerce * 'single-float) +;;; 0.0 +;;; aah +;;; "oops" +;;; yep +;;; tried that on clisp from fink: +;;; [1]> least-positive-short-float +;;; 2.93874s-39 +;;; [2]> (coerce * 'single-float) +;;; *** - floating point underflow +;;; yeah +;;; shall i not try to build sbcl with that? +;;; if you turn off underflow traps, then you get 0.0 +;;; well, that makes sense, i guess +;;; #+clisp +;;; (ext:without-package-lock ("SYSTEM") +;;; (setf system::*inhibit-floating-point-underflow* t)) +;;; (in src/cold/ansify.lisp) #+clisp -(locally - - (in-package "COMMON-LISP") - - ;; no longer needed in CLISP 1999-01-08, hurrah! - #| - ;; ANSI specifies that package LISP defines the type BOOLEAN, and the CMU CL - ;; compiler uses it a lot. This should be trivial to patch in CLISP, except - ;; that CLISP defines FFI:BOOLEAN, which conflicts. Gads.. Here we try to fix - ;; it with some package hacking. (Please do not take this as an example of - ;; good package hacking, I just messed with it until it seemed to work well - ;; enough to bootstrap CMU CL, because I'm highly unmotivated to make elegant - ;; fixes for nonstandard behavior. -- WHN) - (shadow 'ffi:boolean "FFI") - (deftype cl::boolean () '(member t nil)) - (export 'boolean "LISP") - |# +(ext:without-package-lock ("SYSTEM") + (setf system::*inhibit-floating-point-underflow* t)) + +;;;; CMU CL issues - ;; I gave up on using CLISP-1999-01-08 as a cross-compilation host because of - ;; problems that I don't have workarounds for: - (error "can't use CLISP -- no MAKE-LOAD-FORM") - (error "can't use CLISP -- no (FUNCTION (SETF SYMBOL-FUNCTION))") - ) - -;;; CMU CL, at least as of 18b, doesn't support PRINT-OBJECT. In particular, it -;;; refuses to compile :PRINT-OBJECT options to DEFSTRUCT, so we need to -;;; conditionalize such options on the :NO-ANSI-PRINT-OBJECT feature in order -;;; to get the code to compile. (It also fails to do anything useful with -;;; DEFMETHOD PRINT-OBJECT, but that doesn't matter much, since it doesn't stop -;;; the cross-compiler from working.) -;;; -;;; FIXME: SBCL 0.5.0 doesn't support PRINT-OBJECT either. SBCL 0.6.0 will, -;;; at which time this conditional should go away. +;;; CMU CL, at least as of 18b, doesn't support PRINT-OBJECT. In +;;; particular, it refuses to compile :PRINT-OBJECT options to +;;; DEFSTRUCT, so we need to conditionalize such options on the +;;; :NO-ANSI-PRINT-OBJECT feature in order to get the code to compile. +;;; (It also fails to do anything useful with DEFMETHOD PRINT-OBJECT, +;;; but that doesn't matter much, since it doesn't stop the +;;; cross-compiler from working.) #+cmu (progn (warn "CMU CL doesn't support the :PRINT-OBJECT option to DEFSTRUCT.~%") @@ -62,66 +63,98 @@ ;;; /(READ-SEQUENCE *BUFFER* S :START 0 :END 3000)=3000 ;;; /(READ-SEQUENCE *BUFFER* S :START 0 :END 15000)=1096 ;;; /(READ-SEQUENCE *BUFFER* S :START 0 :END 15000)=0 -#+cmu ; FIXME: Remove SBCL once we've patched READ-SEQUENCE. +#+cmu (progn (warn "CMU CL has a broken implementation of READ-SEQUENCE.") (pushnew :no-ansi-read-sequence *features*)) -;;; Do the exports of COMMON-LISP conform to the standard? If not, try to make -;;; them conform. (Of course, ANSI says that bashing symbols in the COMMON-LISP -;;; package like this is undefined, but then if the host Common Lisp were ANSI, -;;; we wouldn't be doing this, now would we? "One dirty unportable hack -;;; deserves another.":-) +;;; This is apparently quite old, according to +;;; : +;;; (error "CMUCL on Alpha can't read floats in the format \"1.0l0\". +;;; the warning relates to a random vinary produced from cvs of +;;; around feb 2000, the corresponding sources to which I never found +;;; (But it seems harmless to leave it here forever just in case.) +#+(and cmu alpha) +(unless (ignore-errors (read-from-string "1.0l0")) + (error "CMUCL on Alpha can't read floats in the format \"1.0l0\". Patch your core file~%~%")) + +#+(and cmu sparc) +(ext:set-floating-point-modes :traps '(:overflow :invalid :divide-by-zero)) + +;;;; OpenMCL issues + +;;; This issue in OpenMCL led to some SBCL bug reports ca. late 2003. +#+openmcl +(unless (ignore-errors (funcall (constantly t) 1 2 3)) + (error "please find a binary that understands CONSTANTLY to build from")) + +;;;; general non-ANSI-ness + +(in-package :sb-cold) + +(defmacro munging-cl-package (&body body) + #-clisp `(progn ,@body) + #+clisp `(ext:without-package-lock ("CL") + ,@body)) + +;;; Do the exports of COMMON-LISP conform to the standard? If not, try +;;; to make them conform. (Of course, ANSI says that bashing symbols +;;; in the COMMON-LISP package like this is undefined, but then if the +;;; host Common Lisp were ANSI, we wouldn't be doing this, now would +;;; we? "One dirty unportable hack deserves another.":-) (let ((standard-ht (make-hash-table :test 'equal)) (host-ht (make-hash-table :test 'equal)) - (cl (find-package "COMMON-LISP"))) + (cl (find-package "COMMON-LISP"))) (do-external-symbols (i cl) (setf (gethash (symbol-name i) host-ht) t)) (dolist (i (read-from-file "common-lisp-exports.lisp-expr")) (setf (gethash i standard-ht) t)) (maphash (lambda (key value) - (declare (ignore value)) - (unless (gethash key standard-ht) - (warn "removing non-ANSI export from package CL: ~S" key) - (unexport (intern key cl) cl))) - host-ht) + (declare (ignore value)) + (unless (gethash key standard-ht) + (warn "removing non-ANSI export from package CL: ~S" key) + (munging-cl-package + (unexport (intern key cl) cl)))) + host-ht) (maphash (lambda (key value) - (declare (ignore value)) - (unless (gethash key host-ht) - (warn "adding required-by-ANSI export to package CL: ~S" key) - (export (intern key cl) cl)) - ;; FIXME: My righteous indignation below was misplaced. ANSI sez - ;; (in 11.1.2.1, "The COMMON-LISP Package") that it's OK for - ;; COMMON-LISP things to have their home packages elsewhere. - ;; For now, the hack below works, but it's not good to rely - ;; on this nonstandardness. Ergo, I should fix things so that even - ;; when the cross-compilation host COMMON-LISP package has - ;; symbols with home packages elsewhere, genesis dumps out - ;; the correct stuff. (For each symbol dumped, check whether it's - ;; exported from COMMON-LISP, and if so, dump it as though its - ;; home package is COMMON-LISP regardless of whether it actually - ;; is. I think..) - ;; - ;; X CMU CL, at least the Debian versions ca. 2.4.9 that I'm - ;; X using as I write this, plays a sneaky trick on us by - ;; X putting DEBUG and FLOATING-POINT-INEXACT in the - ;; X EXTENSIONS package, then IMPORTing them into - ;; X COMMON-LISP, then reEXPORTing them from COMMON-LISP. - ;; X This leaves their home packages bogusly set to - ;; X EXTENSIONS, which confuses genesis into thinking that - ;; X the CMU CL EXTENSIONS package has to be dumped into the - ;; X target SBCL. (perhaps a last-ditch survival strategy - ;; X for the CMU CL "nooo! don't bootstrap from scratch!" - ;; X meme?) As far as I can see, there's no even slightly - ;; X portable way to undo the damage, so we'll play the "one - ;; X dirty unportable hack deserves another" game, only even - ;; X dirtierly and more unportably than before.. - #+cmu - (let ((symbol (intern key cl))) - (unless (eq (symbol-package symbol) cl) - (warn "using low-level hack to move ~S from ~S to ~S" - symbol - (symbol-package symbol) - cl) - (kernel:%set-symbol-package symbol cl)))) - standard-ht)) + (declare (ignore value)) + (unless (gethash key host-ht) + (warn "adding required-by-ANSI export to package CL: ~S" key) + (munging-cl-package + (export (intern key cl) cl))) + + ;; FIXME: My righteous indignation below was misplaced. ANSI sez + ;; (in 11.1.2.1, "The COMMON-LISP Package") that it's OK for + ;; COMMON-LISP things to have their home packages elsewhere. + ;; For now, the hack below works, but it's not good to rely + ;; on this nonstandardness. Ergo, I should fix things so that even + ;; when the cross-compilation host COMMON-LISP package has + ;; symbols with home packages elsewhere, genesis dumps out + ;; the correct stuff. (For each symbol dumped, check whether it's + ;; exported from COMMON-LISP, and if so, dump it as though its + ;; home package is COMMON-LISP regardless of whether it actually + ;; is. I think..) + ;; + ;; X CMU CL, at least the Debian versions ca. 2.4.9 that I'm + ;; X using as I write this, plays a sneaky trick on us by + ;; X putting DEBUG and FLOATING-POINT-INEXACT in the + ;; X EXTENSIONS package, then IMPORTing them into + ;; X COMMON-LISP, then reEXPORTing them from COMMON-LISP. + ;; X This leaves their home packages bogusly set to + ;; X EXTENSIONS, which confuses genesis into thinking that + ;; X the CMU CL EXTENSIONS package has to be dumped into the + ;; X target SBCL. (perhaps a last-ditch survival strategy + ;; X for the CMU CL "nooo! don't bootstrap from scratch!" + ;; X meme?) As far as I can see, there's no even slightly + ;; X portable way to undo the damage, so we'll play the "one + ;; X dirty unportable hack deserves another" game, only even + ;; X dirtierly and more unportably than before.. + #+cmu + (let ((symbol (intern key cl))) + (unless (eq (symbol-package symbol) cl) + (warn "using low-level hack to move ~S from ~S to ~S" + symbol + (symbol-package symbol) + cl) + (kernel:%set-symbol-package symbol cl)))) + standard-ht))