![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to Changes CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [MW Coherent from dump] / coherent / b / bin / c|
Return to Changes CVS log | Up to [MW Coherent from dump] / coherent / b / bin / c |
1.1 root 1: Steve's cc386 compiler change log
2: Begun 1/4/93, earlier logs in Changes.1992.
3:
4: Tue Jan 5 12:14:38 1993 PST
5: n0/gdecl.c
6: In gdecl(), s==LPAREN, deleted "aok = 0;" line.
7: nigel complained that e.g.
8: int (f)(arg) int arg; {}
9: confused the parser (because of the parens around the "f").
10: This line looks odd to me, but nigel had a good reason for
11: wanting it to work, and more to the point it is perfectly
12: acceptible syntax for both K&R and ANSI.
13: Also trivial cosmetic changes.
14:
15: Tue Jan 5 17:35:34 1993 PST
16: n1/i386/tables/leaves.t
17: V4.0.N: added PNDP to code table entry to load addressible
18: doubles in argument context, so that e.g. fn(d) generates
19: push d+4
20: push d
21: for addressible d rather than
22: fldl d
23: call _dp87
24: as before.
25:
26: Tue Jan 5 21:08:45 1993 PST
27: coh/cc.c
28: In runld(), added isvariant(VNDP) code to look for rts
29: in /lib/ndp and pass -L/lib/ndp to ld, as suggested by cef.
30: Before, e.g. "cc -VNDP a.c -f" would execute
31: /bin/ld -X -o a /lib/crts0.o -u _dtefg a.o -lc
32: Now, when -VNDP is specified, it looks for the rts in /lib/ndp
33: and passes -L/lib/ndp to ld, as in:
34: /bin/ld -X -L/lib/ndp -o a /lib/ndp/crts0.o -u _dtefg a.o -lc
35: and you get the right rts and library when you link.
36:
37: Tue Jan 26 09:50:55 1993 PST
38: coh/cc.c
39: Added -L/usr/lib/ndp to ld invocation if -VNDP.
40: n1/i386/tables/leaves.t
41: Added "fwait" at beginning of double push entry (cf. 1/5/93
42: change above), to completion of pending NDP operations.
43:
44: Thu Jan 28 10:27:00 1993 PST
45: cpp/cpp.c
46: Added -VCPLUS option for Comeau. Modified usage() message.
47: Added _addargs() call #if COHERENT to expand CPPHEAD, CPPTAIL.
48:
49: Tue Feb 9 10:47:51 1993 PST
50: coh/cc.c
51: common/diag.c
52: V4.1.1: moved diagnostic output from stdout to stderr.
53:
54: Wed Feb 10 16:19:54 1993 PST
55: coh/cc.c
56: common/diag.c
57: h/var.h
58: V4.1.2: deleted unimplemented VCALLS option, moved VLIB,
59: added VNOWARN option to suppress warning and strict messages.
60:
61: Thu Feb 11 13:30:40 1993 PST
62: coh/Makefile
63: Changed "cc" to "$(CC)" for gcc compilation.
64: common/bput.c
65: Changed "char *cp;" to "unsigned char *cp;" because gcc
66: complained about pointer subtraction char * minus uchar *.
67: n1/out.c
68: Added cast because of gcc complaint about pointer subtraction,
69: as above.
70: n2/i386/outcoff.c
71: In write_sym(), initialize n_type and n_numaux entries
72: appropriately. The gcc-compiled version was writing garbage
73: in symbol entries because these initializations were missing;
74: the cc-compiled version must have worked by accident.
75:
76: Fri Feb 26 09:58:09 1993 PST
77: h/i386/cppmch.h
78: n0/cc0.c
79: V4.1.3: at nigel's request, for ISO compliance, cpp now defines
80: not only COHERENT, MWC, _IEEE and _I386, but also
81: __COHERENT__, __MWC__, __IEEE__, and __I386__.
82:
83: Mon Mar 1 08:53:08 1993 PST
84: n1/i386/tables/div.t
85: n1/i386/tables/mul.t
86: n1/i386/tables/rem.t
87: V4.1.4: User Alexander Shendi reported that the line
88: cp = (char *)(((int) cp1) % ((int) cp2));
89: gave an internal error, actually a no match on the '%';
90: The code tables expected the result of the '%'
91: to be LONG (int or uint), and the (char *) cast on the '%' node
92: caused the type to mismatch. The same code also failed for '*'
93: and '/'. The code sucks, but it is legal and there's no
94: reason why it should fatal the compiler. Changed the node
95: type on appropriate code table entries from "LONG" to "DWORD"
96: and now it works.
97:
98: Tue Mar 2 17:54:00 1993 PST
99: n0/cc0.c
100: V4.1.5: hal complained that
101: typedef foo baz;
102: where foo is not defined gave
103: 1: a.c: missing semicolon
104: which is none too clear. Now it says
105: 1: a.c: illegal typedef
106: instead.
107:
108: Wed Mar 3 17:45:30 1993 PST
109: n0/expr.c
110: V4.1.6: hal complained about the strict warning produced by:
111: struct s1 { int s_i; } s1;
112: struct { int s2_i; } s2;
113: f() { s2.s_i = s1.s_i; }
114: which was
115: 3: b.c: Strict: structure ".1" does not contain member "s_i"
116: where the ".1" is the compiler-generated tag for the untagged
117: structure. Now the warning message says
118: 3: b.c: Strict: structure does not contain member "s_i"
119: instead.
120:
121: Wed Mar 24 09:25:43 1993 PST
122: n2/i386/outcoff.c
123: V4.1.7: changed the code which constructs derived type info
124: for DT_ARY, DT_PTR and DT_FCN in keeping with <coff.h> change.
125:
126: Thu Apr 1 15:31:33 1993 PST
127: n3/i386/igen.c
128: Write "\t.intelorder\n" near start of assembler output file.
129:
130: Wed Apr 7 22:02:30 1993 PDT
131: n0/gdecl.c
132: V4.1.8: In genum(), changed
133: deflookup(SL_VAR, LL_EXT);
134: to
135: deflookup(SL_VAR, ll);
136: to enter enumeration members in the symbol table
137: at the current lexical level, not at external level.
138: User H. P. Heidinger reported that e.g.
139: f() { enum e { e1 } fe; }
140: g() { enum e { e0, e1 } ge; }
141: gave a bogus redeclaration error, because the local enum
142: members were entered at the wrong lexical level.
143:
144: Thu Apr 8 14:54:51 1993 PDT
145: n0/init.c
146: V4.1.9: User H. P. Heidinger complained (as many have before)
147: that bitfield initialization was not supported; e.g.
148: struct s {
149: int s_bit0:1;
150: int s_bit1:1;
151: } s = { 1, 1 };
152: previously got a "cannot initialize bitfields" error
153: from the parser. Added "bfinit()", now it works (I hope!).
154:
155: Mon Apr 12 11:47:33 1993 PDT
156: V4.2.0: Changed constant folder to fold unsigned constants
157: correctly and to fold double constants.
158: h/i386/mch.h
159: Added compilation option FOLD_DOUBLES.
160: This assumes host and target use same fp representation.
161: Added typedefs for uival_t and ulval_t, the unsigned types
162: corresponding to ival_t and lval_t.
163: h/cc0.h
164: Recoded opdope table constants from octal to hex.
165: Added new constant FOLD, indicating op can be folded.
166: Separated extern function definitions from variables.
167: Added dval_to_d() and d_to_dval, defined in bind.c.
168: n0/i386/bind.c
169: Added #if FOLD_DOUBLES code for dval_to_d() and d_to_dval(),
170: reversing byte order to convert host double to/from DCON.
171: n0/dope.c
172: Added FOLD to opdope for operations which support constant
173: folding; this allows fold0() to be much more efficient.
174: n0/fold.c
175: Extensive changes to handle typed constant folding (including
176: double constant folding). Previously, the constant folder
177: understood int and long constants but did not understand
178: unsigned or double constants. This meant
179: (1) it sometimes folded unsigned constants incorrectly, and
180: (2) it could not compile programs with constant double
181: expressions in initializers.
182: The new double constant folding code uses the host fp
183: arithmetic, but it is conditionalized #if FOLD_DOUBLES so that
184: it can be excluded for cross-compilation if desired.
185:
186: Mon Apr 12 13:32:45 1993 PDT
187: n1/mtree1.c
188: Cleaned up comments and formatting, no substantive changes.
189:
190: Mon Apr 12 15:36:54 1993 PDT
191: n1/mtree1.c
192: Rewrote fold1() for clarity. The handling of typed constants
193: here needs to be fixed up like n0/fold.c/fold0() above.
194: n1/i386/tables/aadd.t
195: V4.2.1: Changed the right argument types for two entries
196: in the += code table from LONG to DWORD to fix a "no match"
197: bug reported by H. P. Heidinger:
198: cp += (int)buf;
199: previously no matched for auto char buf[1].
200: And why, you might ask, should the right type be DWORD
201: rather than LONG, i.e. why should allowing pointers on the
202: right matter in this case, where the right is not pointer?
203: The parser, knowing that T_PTR and T_INT are bit compatible,
204: changes the "&" node type it outputs from PTR to S32
205: because of the cast and then writes the right arg as
206: &@ S32
207: Aid S32 at -12
208: (actually with a scaling multiply by 1 around it).
209: The automatic then gets expanded, resulting in
210: &@ S32
211: *@ S32 flag=3000
212: + Ptr flag=800
213: Reg Ptr flag=400 %ebp
214: Icon S32 flag=a8 -12
215: and the & * foo gets simplified to foo, leaving
216: + S32 flag=800
217: Reg Ptr flag=400 %ebp
218: Icon S32 flag=a8 -12
219: So far so good, the type of the add node is still S32. But
220: then the associative expression folder in n1/mtree1.c/fold1()
221: looks at this and adjusts the top type, as it must for
222: additions involving ints and pointers which have been
223: rearranged; it adjusts the ADD node type to
224: + Ptr flag=800
225: Reg Ptr flag=400 %ebp
226: Icon S32 flag=a8 -12
227: and this results in the "no match" in aadd.t because the
228: right type is inappropriate. Changing the right types in
229: aadd.t fixes the problem.
230:
231: Mon Apr 12 17:08:22 1993 PDT
232: n1/i386/tables/asub.t
233: V4.2.2: identical change to above for -= instead of +=.
234:
235: Thu Apr 15 12:06:35 1993 PDT
236: coh/cc.c
237: V4.2.3: Jeff Moskow ([email protected], 1-508-448-3959)
238: reported a bug which I traced to cc.c.
239: cc -DSTRING="\"\
240: 12345678901234567890123456789012345678901234567890\
241: <nine more lines identical to the above>
242: \"" a.c
243: caused odd behavior. The driver was copying each cpp arg
244: into a static (i.e. in .data) 64-character buffer without
245: bounds checking, eventually overwriting the static data
246: used to initialize the keyword table. Simplified the way
247: it works (the cpp arg copy was usually unnecessary!)
248: and added appropriate bounds checking. Finally figured out
249: the hacky code which allows runpp() and runld() to grovel
250: in the args when a single cc arg contains multiple arg types,
251: after looking at it many times before. Changed the _I386
252: default for "-r" compilation from "l.out" to "a.out".
253: Added usage() as separate routine, mentioning -? option.
254: Added -? option to print verbose usage information.
255: Cleaned up some comments and added some others.
256:
257: Thu Apr 15 12:53:29 1993 PDT
258: coh/cc.c
259: Added a few comments. Deleted -k from verbose usage message,
260: since neither ld286 nor ld386 understands it.
261:
262: Fri Apr 16 12:13:08 1993 PDT
263: n0/expr.c
264: V4.2.4: In build(), case QUEST, added code to check if the
265: condition type is computational and insert a conversion node
266: if not. A bug reported by Jeff Moscow included code of the form
267: a = ((short)b) ? c : d;
268: While the noncomputational condition type did not appear to
269: cause any problems, it looks better if the type is correct.
270: n1/mtree0.c
271: For case CONVERSION/CAST, after the modtree(lp, ...) call,
272: check to see if the modified lp op is COMMA or QUEST;
273: the code above which distributes the conversion over the
274: COMMA or QUEST needs to be executed again in this case,
275: e.g. for (type1)(type2)((a)?b:c). At the same time, added
276: a test for (type1)(type2)type1 where type2 is wider,
277: in which case both the conversions are irrelevant.
278: Other minor nonsubstantive changes.
279: n1/i386/table0.c
280: Added clarifying comments, no substantive changes.
281: n1/i386/tables/leaves.t
282: Added comment about the asm.c kludge below to the appropriate
283: code table entry. No substantive changes.
284: n2/i386/asm.c
285: In assemble(), case OF_MOV, added code to kludge e.g.
286: movw %ax, %edi
287: to
288: movw %ax, %di
289: just like the similarly disgraceful kludge for ZMOV[SZ]W.
290: This fixes (after a fashion) a bug reported by Jeff Moscow:
291: for short s and int struct member i, his example was
292: a = (short)((((short)(p->i)) ? ((short)(p->i)) : (2)));
293: after macro expansion, but the much simpler
294: s = (short)(int)(short)i;
295: has the same bug, namely a fatal aerr() on the "movw" above.
296: Spent a lot of time trying to fix it cleanly, but
297: without success. The fix includes the similar char case (movb),
298: but I could not make that aerr() happen, the difference is
299: somehow related to the PBYTE and bytereg() code.
300: n2/i386/outcoff.c
301: In setrefnum(), added ignore flag to allow compilation -VTPROF
302: to work without -S or -g. The output writer now ignores the
303: table profiling debug items in this case.
304:
305: Mon Apr 19 14:25:47 1993 PDT
306: n2/i386/asm.c
307: V4.2.5: fixed kludge of 4/16/93, the previous implementation
308: was seriously buggy, breaking byte and word mov instructions.
309:
310: Thu Apr 22 00:26:55 1993 PDT
311: coh/cc.c
312: V4.2.6: added global "int monolithic = 1;" #if MONOLITHIC.
313: The -tBNM options reset this and tempsize to 0 to tell the
314: driver to use phases instead of monolithic compilation and
315: to use disk temps instead of memory temps. Changed
316: #if MONOLITHIC conditionalized code accordingly.
317:
318: Fri Apr 23 16:30:30 1993 PDT
319: n0/sharp.c
320: V4.2.7: In getfname(), added code to expand ID in filename,
321: so that e.g.
322: #define FOO <stdio.h>
323: #include FOO
324: works as expected.
325:
326: Wed May 5 08:52:24 1993 PDT
327: n0/cc0.c
328: n0/gdecl.c
329: V4.2.8: The 1/5/93 change above to allow the parser to
330: understand declarators like "int (f)(arg) int arg; {}"
331: screwed up the offsets assigned to parameters
332: of functions returning pointers to functions,
333: as recently reported by Harry Pulley.
334: For example, in
335: int (*fpfi(a, b))() int a, b; { ... }
336: the compiler generated incorrect code for references
337: to a and b in fpfi. In cc0.c/xdef(), now set "nargs = 0;"
338: before calling gdecl(). In gdecl(), set argok to 1 only
339: if nargs==0. After the recursive call to gdecl() for LPAREN,
340: reset argok to 0 if the recursive call set nargs != 0.
341:
342: Fri May 14 13:20:27 1993 PDT
343: n0/expr.c
344: V4.3.0: In build(), case QUEST, the change of V4.2.4 on 4/16/93
345: for ?: with noncomputational type in the condition was wrong
346: for objects with nonempty DIM lists.
347: This fixes a bug reported today by Jeff Moskow: e.g.
348: localjeff = jeff ? jeff : b;
349: where all the objects are pointers to functions returning int
350: generated a byte rather than long compare.
351:
352: Thu Jun 3 17:38:09 1993 PDT
353: coh/cc.c
354: V4.3.1: Added -VVERSION number to print version number
355: to standard error.
356:
357: Wed Jul 7 10:58:50 1993 PDT
358: n1/i386/mtree2.c
359: User Douglas J. Toltzman <[email protected]> reported a
360: bug with ushort (also char, uchar, short) bitfield assignments.
361: In the field extraction tree modifier modefld():
362: deleted code which made tree top noncomputational,
363: changed the shift widths from assuming byte or word width
364: to always assuming dword,
365: added code to always treat field width 1 as unsigned,
366: added clarifying comments.
367:
368: Wed Jul 7 13:56:31 1993 PDT
369: n2/i386/outcoff.c
370: User Gabor Zahemsky <[email protected]> reported a bug where -g
371: caused a cbotch() in outdloc(), basically because a #line
372: directive made the label at the start of a function fall at
373: line 0 of the function. The output writer now distinguishes
374: function vs. line number LNNUM entries by a new flag field
375: rather than by the line number being 0 or nonzero.
376:
377: Wed Jul 7 20:39:06 1993 PDT
378: coh/cc.c
379: h/var.h
380: Added -VREADONLY option to recognize "readonly" keyword, after
381: nigel complained about it allowing "readonly" by default.
382: Now it recognizes "readonly" only if -VREADONLY.
383:
384: Thu Jul 8 12:05:22 1993 PDT
385: coh/cc.c
386: h/cc0.h
387: n0/cc0.c
388: n0/cc0key.c
389: n0/expand.c
390: Added predefined cpp macro __BASE_FILE__ at nigel's request.
391: __BASE_FILE__ expands to a string giving the name of the
392: original source file being compiled. This is like the
393: __FILE__ macro, except that within a #included header file
394: __FILE__ expands to the header file name while __BASE_FILE__
395: expands to the original source file name.
396:
397: Wed Jul 14 14:05:00 1993 PDT
398: n0/dope.c
399: n0/fold.c
400: nigel complained that e.g.
401: char a[(1,2)];
402: (that's a ','-operator, result of cpp macro expansion,
403: not a multi-dim. array) would not compile, issuing an
404: "array bound must be a constant" message.
405: The parser's constant folder previously ignored COMMA,
406: so it was always nonconstant; now COMMA with constants on both
407: sides returns the RHS, and the above compiles as expected.
408:
409: Wed Jul 28 13:25:46 1993 PDT
410: n0/cpp.c
411: V4.3.4: In attempting to run Intek C++, Jay discovered that
412: "/lib/cpp -C" malfunctioned on comments containing
413: single or double quotes (' or "). Now cpp in -C mode
414: ignores ' or " within comments; this includes C++-style
415: online //-delimited comments if -VCPLUS is specified.
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.