path: root/Build/source/utils/mtx/mtx-0.60d/libp2c/ChangeLog

Wed Mar 10 15:03:20 1999  Alexander L. Belikoff  <abel@blinky>

	* src/p2c.h: DOUBLE defined as double

Tue Mar 31 14:57:26 1998  Alexander L. Belikoff  <abel@blinky>

	* src/trans.h:
	Magic limits (MININT, MAXINT), causing random p2c bugs on the Alpha replaced
	with LONG-MIN/MAX from limits.h (integers in Pascal include longs in C,
	according to p2c)

Tue Sep 23 09:14:22 1997  Harvey J. Stein  <hjstein@bfr.co.il>

	* src/p2c.proto, src/funcs.c, src/p2c.hdrs:
	Adjustments to get pad to work correctly.

Thu Jul  3 20:41:55 1997  Harvey J. Stein  <hjstein@bfr.co.il>

	* src/p2c.proto, src/sys.p2crc, src/trans.h, src/expr.c, src/funcs.c, src/p2c.hdrs:
	Bug: Consider:

	   procedure foo(x: array[i..j : integer] of varying[k] of char);
	   ...
	   var z : array[3..12] of varying[20] of char;
	   ...
	   foo(z);

	This should become something like:

	   foo(int i, int j, int k, char x[j-i+1][k])
	   ...
	   char z[10][20];
	   ...
	   foo(3, 12, 20, z);

	Instead, it was generating:

	   foo(3, 12, 255, z);

	which totally screws up accessing z.

	Analysis: To do this correctly, p2c has to keep dereferencing z as it
	processes the hidden arguments i, j, & k so that it's working on an
	expression of the appropriate type each time it tries to find either
	the size of the string or the start & end points of arrays.  Instead,
	it just wasn't.

	Fix done: I couldn't figure out exactly how to do the above.  But, p2c
	calls strmax_func to get the length of the character string.  I just
	built some additional dereferencing into it so that it gets the
	correct length in the above sort of situation.  So, changed
	strmax_func in expr.c.


	Bug: Index(x,y) is a common pascal extension for finding a substring
	in a string.  p2c handles it but it wasn't working right on Dec code.

	Analysis:  Some dialects of pascal consider x to be the string and y
	to be the substring.  Others switch x & y.  p2c supports (substring,
	string), and Dec pascal supports the opposite.

	Fix: There was some support for switching index argument order, but it
	was tightly coupled to HP compiler directive support.  I added a
	comment switch ("IndexOrder") for additional control of this feature.
	Changed func_strpos in funcs.c, and added the switch in trans.h.

Fri Jun 13 07:22:26 1997  Harvey J. Stein  <hjstein@blinky.bfr.co.il>

	* src/p2c.hdrs, src/p2c.proto, src/pexpr.c, src/decl.c, src/funcs.c:
	Fixed bug in [truncate] handling - I didn't clear the new Meaning
	fields (fakenumargs & truncated) at meaning creation time, so they
	were sometimes showing up on non-truncated variables.  Changed
	initmeaning (decl.c).

	Moved hidden number_of_args from beginning to end of arg list.  At the
	beginning it was interfering with the return hidden variable,
	causing return to not be generated.  Changes were in p_funcdecl
	(decl.c), func_present (funcs.c), and p_funcarglist (pexpr.c).  Added
	has_truncate function to pexpr.c to check whether an arg list has the
	truncated attribute.

	Argument variables were sometimes getting shadowed just because
	they're initialized.  Problem was in var_reference, which clones
	Meanings that have constdefns.  But arg variables don't need to be
	cloned since the constants are handled by p_funcarglist.  So skipped
	cloning if Meaning's kind is MK_PARAM or MK_VARPARAM.  Previous fix of
	always cloning, but only copying meanings if not MK_PARAM seems to
	have been incorrect.

Sun Jun  1 13:15:06 1997  Harvey J. Stein  <hjstein@blinky.bfr.co.il>

	* src/sys.p2crc, src/trans.h, src/pexpr.c, src/parse.c, src/lex.c, src/p2c.hdrs, src/p2c.proto, src/funcs.c, src/decl.c:
	Enhancements:

	Added support for [truncate].  This involved:
	   1. Added a truncated flag & a fakenumargs flag to the Meaning
	      structure.
	   2. In p_funcdecl (decl.c), recognizing the [truncate] attribute in
	      argument lists.  If it's seen, the argument's truncated flag is
	      set, and it and all arguments following are given default values
	      (assuming that they don't already have default values).  The
	      default values gets p2c to automatically generate the correct
	      number of arguments for calling the subroutine when some are
	      left out.
	   3. In p_funcdecl (decl.c), if any of the args are tagged with
	      [truncate], then add a hidden argument to the front of the
	      argument list - number_of_args.  Flag it with fakenumargs.
	   4. In p_funcarglist (pexpr.c), when the fcn has a fakenumargs
	      parameter as the first parameter, (and truncateisdefalt is zero)
	      compute the index of the last argument supplied, and pass it as
	      the first argument in the call.
	   5. In funcs.c, added func_present to handle the PRESENT(x) calls.
	      It parses the call, figures out the index of the argument in the
	      fcns calling sequence (say n), and generates an expression
	      (number_of_args >= n) to be the output for the PRESENT(x) call.
	   6. Added new config parameter - TruncateIsDefault to turn off
	      PRESENT support (in case one doesn't want the extra arg to be
	      generated).

	Added support for CONTINUE & BREAK.  Just had to basically declare
	them as keywords in lex.c, and recognize them in parse.c, and declare
	their tokens in trans.h (since p2c already supported them internally).

	Added descriptions for SkipReIncludes, NoOutput, and TruncateIsDefault
	to sys.p2crc.

	Bug fixes:

	Bug: args with default values which are also used in local functions
	lose their default values.

	Attempted fix: in var_reference (pexpr.c), don't zero out the
	constdefn when the arg is cloned if the arg is a parameter.

	Problems: This works, but it sometimes causes bogus declarations of
	the old variable to be output.

Wed May 14 14:27:32 1997  Harvey J. Stein  <hjstein@blinky.bfr.co.il>

	* src/stuff.c, src/p2c.hdrs, src/p2c.proto, src/out.c, src/expr.c, p2c.patch:
	(Updated p2c.patch for last set of changes)

	Bug: p2c was seg faulting when given expressions such as:
	   there :=
	   '1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890' +
	   '1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890' +
	   '1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890' +
	   '1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890' +
	   '1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890';

	and:

	   there :=
	   '1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890' +
	   '1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890' +
	   '1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890' +
	   '1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890' + where;

	when there is VARYING [XXX] OF CHAR;

	Analysis: There seem to be all sorts of fixed length character arrays
	around the code.  It seems that especially long expressions and/or
	character strings can cause them to overflow.

	Fix: I upped some of the limits, which at least avoided core dumps on
	the code I'm currently processing, but really one needs to go through
	and properly keep track of the amount copied into strings so as to
	avoid buffer overruns.  But, this is a real pain in the ass because of
	the shitty semantics of the standard C library.

	Basically, the stack allocated ones that I noticed in expr.c and out.c
	I upped to length 2000, and in stuff.c, I upped the size of the
	strings in formatbuf[] to 4000.

Tue May 13 19:27:00 1997  Harvey J. Stein  <hjstein@blinky.bfr.co.il>

	* src/lex.c, src/p2c.hdrs, src/p2c.proto, src/trans.c, src/Makefile, src/decl.c, src/expr.c, src/funcs.c:
	Minor fixes/changes:
	   -Was calling close in handle_include (lex.c) when it should have
	    been fclose.
	   -Modified src/Makefile to allow for using insure on the source code
	    (to catch memory bugs).


	Major fixes:

	Bug: p2c occasionally was freeing memory that was already freed.

	Analysis: This seems to occur sometimes on constant string
	     definitions because mp->val.s might point to the same string as
	     mp->constdefn->val.s, and forget_ctx tries to free them both.

	Fix: In decl.c, in forget_ctx: After calling free_value(&mp->val) when
	     mp->kind == MK_CONST, instead of always calling
	     freexpr(mp->constdefn), only call it when mp->kind != MK_CONST,
	     or when mp->constdefn->val.s != mp->val.s.

	     I also tried to fix this by copying mp->val.s in p_constdecl when
	     creating mp->constdefn->val.s, but never got this to work
	     correctly, so I commented it out.


	Bug: p2c core dumped on:

	     readv(data,         short_string);

	     when data is a packed array[1..7] of char, and
	     short_string is a packed array[1..3] of char.

	Analysis: What was happening was that in handleread_text (funcs.c),
	     when the readv argument is an array, it was checking
	     ord_range(ex->val.type->indextype, ...), and ex was NULL.

	Fix: It seems like this was a typo, so I change it to check
	     var->... instead of ex->...


	Bug: p2c core dumped on:

	     WRITELN('%SQLMOD ', msg);

	     when msg was of type varying[x] of char.

	Analysis: Some part of the code was converting the above string to
	     "%%SQLMOD %s".  It seems that cleansprintf (from expr.c) didn't
	     realize that % is escaped by %%, so it thought that there should
	     be 2 arguments (maybe 3?) for the above printf string instead of
	     one, and thus was going off of the end of the args list (i.e. -
	     ex->args[j] out of range).

	Fix: Inside the for loop, if the next character is also a %, then
	     skip arg processing, and skip the next character (just copy it
	     into fmtbuf).


	Bug: p2c core dumps on:

	     readv(c, i);

	    when c is a char and i is an integer.

	Analysis: In funcs.c, in handleread_text, var is never getting
	     initialized - it's getting passed in as NULL, and it only gets
	     set if isstrread is true.  Since c is a char, fex->val.type->kind
	     is TK_CHAR, not TK_STRING, so isstrread is false, so var fails to
	     get initialized.

	Fix: After the case that var is set when isstrread is true, if it's
	     still NULL, we set it to p_expr(NULL).  This prevents the core
	     dump, but produces incorrect code!  It outputs:

	        fscanf(c, "%ld", &i);

	     when it should more likely be something along the lines of
	     sscanf(&c, ...) (but safer).

Mon May  5 15:23:04 1997  Harvey J. Stein  <hjstein@blinky.bfr.co.il>

	* src/out.c, src/p2c.hdrs, src/p2c.proto, src/trans.h, src/decl.c, src/lex.c, p2c.patch:
	New features (out.c, trans.h):
	   New parameter NoOutput - setting it to 1 stops p2c from generating
	   output.  Useful when trying to simulate Dec style modules via
	   include files - just convert INHERIT to %include, and put
	   {*NoOutput=1} at the top of each include file, and p2c does the
	   right thing.  This was done by not pushing characters into the
	   output stream when nooutput is 1.  The real code changes occur in
	   out.c, in putc_outf, and puts_outf.  Accumulating the output, but
	   turning off writes (i.e. - wiring it into flush_outfilebuf) failed
	   because of timing differences btw when comment directives are
	   processed and when output occurs.

	Bug fixes:

	   Bug: Some variable declarations weren't getting output.  Most
	   noticable were the automatically generated Return variables.

	   Analysis: When using the {INTF-ONLY} directive in include files
	   then the fwd declaration of fcns was being interpreted as the fcn
	   definition, and so the variables generated when the declaration was
	   seen were marked as "wasdeclared".  Then, when the code for the
	   function was being generated, these variables weren't getting
	   output because "wasdeclared" was 1.  This was most visible from the
	   automatically generated Return variable, but occasionally happened
	   to other variables.

	   Fix: in decl.c:activatemeaning - added resetting of "wasdeclared"
	   if we can identify that the variable is the autogenerated return
	   variable.  Detected them by snarfing code from isretvar.  But this
	   was a lousy fix because as mentioned above, there are other
	   variables which are occasionally affected.  And, it's no longer
	   needed because of the new NoOutput feature, so I commented out the
	   change, but left the code there as a reminder of the problem.


	   Bug: Comments weren't getting output in the correct places.  This
	   is a serious problem when using {EMBED comments to insert C code
	   into the output stream.  Sometimes code from the {EMBED comment
	   given in an include file showed up in the includer.  Sometimes the
	   code didn't get emitted at all because of mistiming against
	   {nooutput} directives.

	   Analysis:  p2c builds up a comment string structure as it parses
	   the code, and hangs it off of the code's internal representation.
	   Presumably this is so that when doing code transformations, the
	   comments are output in the correct places.  However, this seems to
	   sometimes fail when using %include files.  It seems that the
	   comments weren't getting processed & output until the current block
	   was finished.  Included files don't start a new block because one
	   could conceivably include random snippets of code anywhere you want
	   (inside of RECORD declarations, for example).  Thus, sometimes
	   comments from inside the include file don't get processed & output
	   until well after the include file itself is finished being
	   processed.

	   Fix: Added the call flushcomments(NULL, -1, -1) to both push_input
	   and pop_input (in lex.c).  This seems to have fixed it with no
	   noticable side effects.

Tue Apr 29 18:36:51 1997  Harvey J. Stein  <hjstein@blinky.bfr.co.il>

	* src/p2c.hdrs, src/p2c.proto, src/lex.c:
	Added recognition of AND_THEN and OR_ELSE.

	* src/trans.h, src/lex.c, src/p2c.hdrs, src/p2c.proto:
	Added tracking of files that have been included so that they can
	optionally be ignored the second time they're seen.  This is needed
	because a module (say X) can inherit from two other modules (say Y and
	Z), both of whom inherit from a fourth module (say Z).  Converting the
	inherits to includes thus gets Z included twice.  If it's not ignored,
	it causes p2c to start renaming procedures so as to avoid name
	clashes.

	Added an rc file parameter (skipreincludes) for controling the above
	action.

	* src/decl.c, src/p2c.hdrs, src/p2c.proto, src/parse.c:
	Various adjustments & bug fixes suggested by Dave Gillespie:

	decl.c:  Call format_s so that we work with a copy of the string
	         when removing dollar signs - a little more robust.

	         Missing ! on strcicmp caused bogus externs to be emitted on
	         some variable declarations.

	parse.c: Sprinkle afew p_attributes around p_block so that p2c
	         properly parses attributes in front of declaration blocks.
	         In particular, this fixes the problem where p2c was choking
	         on [hidden] followed by TYPE, which appears in analytic.pas.

Wed Apr 23 12:17:08 1997  Harvey J. Stein  <hjstein@blinky.bfr.co.il>

	* src/p2c.proto, src/decl.c, src/p2c.hdrs:
	Modified addfield so that it removes dollar signs from record field
	identifiers.  This is more consistent with the handling of other
	identifiers, and also needed by much of the Pascal code.

Mon Apr 21 10:41:53 1997  Harvey J. Stein  <hjstein@blinky.bfr.co.il>

	* p2c.patch: Submitted new patch to Dave G.

	* src/p2c.proto, src/lex.c, src/p2c.cat, src/p2c.hdrs, src/Makefile:
	Changed VALUE from a keyword to a possible keyword when working with
	Dec Pascal code.  Original code set it to be a keyword for Dec, and a
	possible keyword otherwise.  However, if it's a keyword, then p2c
	doesn't accept it the name of a slot in a record.  Since Dec accepts
	this usage, I figured that VALUE should always be a possible keyword
	instead of a keyword.

Thu Apr 17 11:33:22 1997  Harvey J. Stein  <hjstein@blinky.bfr.co.il>

	* p2c.patch: patches generated & mailed to Dave Gillespie

Wed Apr 16 16:26:07 1997  Harvey J. Stein  <hjstein@blinky.bfr.co.il>

	* src/p2c.hdrs, src/p2c.proto, .cvsignore, examples/.cvsignore, src/.cvsignore, src/decl.c, src/p2c.cat:
	Parser didn't expect a ( after the OTHERWISE keyword.  Modified
	decl.c to expect it.

	* src/INSTALL, src/citmods.c, src/comment.c, src/decl.c, src/dir.c, src/expr.c, src/funcs.c, src/hpmods.c, src/lex.c, src/loc.p2clib.c, src/loc.p2crc, src/makeproto.c, src/out.c, src/p2c.cat, src/p2c.h, src/p2c.hdrs, src/p2c.proto, src/p2cc.cat, src/p2cc.man, src/p2cc.perl, src/p2clib.c, src/parse.c, src/pexpr.c, src/string.pas, src/system.imp, src/system.m2, src/trans.h, src/turbo.imp:
	Import of p2c 1.21alpha2

	* src/INSTALL, src/citmods.c, src/comment.c, src/decl.c, src/dir.c, src/expr.c, src/funcs.c, src/hpmods.c, src/lex.c, src/loc.p2clib.c, src/loc.p2crc, src/makeproto.c, src/out.c, src/p2c.cat, src/p2c.h, src/p2c.hdrs, src/p2c.proto, src/p2cc.cat, src/p2cc.man, src/p2cc.perl, src/p2clib.c, src/parse.c, src/pexpr.c, src/string.pas, src/system.imp, src/system.m2, src/trans.h, src/turbo.imp:
	Initial revision

	* HP/import/asm.imp, HP/import/fs.imp, HP/import/general_0.imp, HP/import/general_1.imp, HP/import/general_2.imp, HP/import/general_3.imp, HP/import/general_4.imp, HP/import/hpib_0.imp, HP/import/hpib_1.imp, HP/import/hpib_2.imp, HP/import/hpib_3.imp, HP/import/iodecl.imp, HP/import/misc.imp, HP/import/rnd.imp, HP/import/sysdevs.imp, HP/import/sysglobals.imp, HP/include/asm.h, HP/include/fs.h, HP/include/general_0.h, HP/include/general_1.h, HP/include/general_2.h, HP/include/general_3.h, HP/include/general_4.h, HP/include/hpib_0.h, HP/include/hpib_1.h, HP/include/hpib_2.h, HP/include/hpib_3.h, HP/include/iodecl.h, HP/include/misc.h, HP/include/rnd.h, HP/include/sysdevs.h, HP/include/sysglobals.h, Makefile, README, examples/Makefile, examples/README, examples/basic.doc, examples/basic.p, examples/c/basic.c, examples/c/cref.c, examples/c/e.c, examples/c/fact.c, examples/c/self.c, examples/cref.p, examples/e.p, examples/fact.p, examples/p2crc, examples/self.p, ftp.README, src/COPYING, src/HISTORY, src/Makefile, src/NOTES, src/README, src/p2c.man, src/stuff.c, src/sys.p2crc, src/trans.c:
	Import of p2c 1.21alpha2

	* HP/import/asm.imp, HP/import/fs.imp, HP/import/general_0.imp, HP/import/general_1.imp, HP/import/general_2.imp, HP/import/general_3.imp, HP/import/general_4.imp, HP/import/hpib_0.imp, HP/import/hpib_1.imp, HP/import/hpib_2.imp, HP/import/hpib_3.imp, HP/import/iodecl.imp, HP/import/misc.imp, HP/import/rnd.imp, HP/import/sysdevs.imp, HP/import/sysglobals.imp, HP/include/asm.h, HP/include/fs.h, HP/include/general_0.h, HP/include/general_1.h, HP/include/general_2.h, HP/include/general_3.h, HP/include/general_4.h, HP/include/hpib_0.h, HP/include/hpib_1.h, HP/include/hpib_2.h, HP/include/hpib_3.h, HP/include/iodecl.h, HP/include/misc.h, HP/include/rnd.h, HP/include/sysdevs.h, HP/include/sysglobals.h, Makefile, README, examples/Makefile, examples/README, examples/basic.doc, examples/basic.p, examples/c/basic.c, examples/c/cref.c, examples/c/e.c, examples/c/fact.c, examples/c/self.c, examples/cref.p, examples/e.p, examples/fact.p, examples/p2crc, examples/self.p, ftp.README, src/COPYING, src/HISTORY, src/Makefile, src/NOTES, src/README, src/p2c.man, src/stuff.c, src/sys.p2crc, src/trans.c:
	Initial revision