(Message inbox:32)
Date:          Mon, 03 Jul 89 21:15:32 CET
From:  V61%DHDURZ1.BITNET@lbl.gov
Subject:       Flex, bug fix, improvments, patches for Minix & TOS
To:  vern@lbl-csam.arpa

At first I have to thank you for your wonderful program. I had ported the
old version to OS9,TOS (Atari ST) and Minix and the new version 2.1 Beta
to Minix and TOS.
     
While porting and using flex I detected a bug and made some improvements.
I have included a  shared, compressed and uuencoded file contaning all cdiffs
and additional files (Sorry, but I'm on EBCDIC-Bitnet) and a short discussion
of the changes. Even some of the TOS specific changes might be of general
interest !
     
I posted these cdiffs to the minix discussion group, but I think it's up
to you to post them to the unix-sources group. If you plan to post even
the TOS compiler specific patches please contact me because there might be
further compiler (P.D.) additions. If you have an interest I could also
port the new version to OS9 -- this is a little bit more difficult, because
OS9 uses CR as end of line character (the EOL char. is coded into the
initscan.c tables,...). It is necessary to change all occurences of '\n' to
macros and variables and it's useful to add a new -n options (see commented
line in main.c)
     
     
     
The changes:                                                  (1.7.89 RAL)
     
 - Bug fix: The original flex didn't like trailing spaces in exclusive start
     condition lists ! If you add an trailing space to line 68 in scan.l
     
     "%x SECT2 SECT2PROLOG SECT3 CODEBLOCK PICKUPDEF SC CARETISBOL NUM QUOTE "
     
     you get a misleading error message:
     
     "Syntax error at line 69: bad start condition list"
     
     This bug can either be fixed in parse.y or in scan.l . I have chosen the
     last because there the fix is minimal: Just change the rule (line 128)
     
      "\n"    to   "{OPTWS}\n"
     
 - Enhancements:
    - new option "-?" that provides some help information about the other
      flags (main.c)
    - new option "-aTMPPATH" that allows a redefinition of the standard
      path for the temporary file (this might be rather large >200k if
      F,f options are selected). (main.c, flexdef.h (l.376))
    - hexdump of illegal characters -- this proved to be a useful debugging
      tool especialy if invisible control characters occur which weren't
      covered by the rules. (scan.l  fprintf statement line 129,...)
     
 - Patches due to TOS
    - General: TOS uses CR,LF as end of line char., Flex wants only a single
      LF as EOL char. Therefore all I/O must be translated using f* calls.
      This is done everywhere besides the YY_INPUT macro (flex.skel (scan.c),
      line 31) that uses a low level 'read'. This should be definitly changed
      to fread, so that all I/O calls occur on the same level.
    - the short action_file_name has been "flexXXXXXX.tmp", but that's too
      much for TOS,MSDOS ! I changed it to "flexXXXX.tmp" in main.c (patch
      covered by the -a option additions)
    - some compilers don't like external names that are ambiguous within
      the first 8 characters. I defined macros that change all these long
      external names to names that are unique within the first 8 characters.
      Just define SHORT_EXTERNAL_NAMES to use this feature (flexdef.h)
    - some statement changes that some compiler don't like:
      typedef *xxx[]  -> typedef **xxx       (flexdef.h.,l.308)
      "/*" -> '/','*' within a comment in (scan.l, l.209)
    - changed short "lexyy.c" to "lex_yy.c" what's more similar to the unix
      "lex.yy.c" (main.c).
    - a few further really compiler dependent changes provided with
      #ifdef ATARI && LATTICE res. TURBO braces.
     
 - Additional TOS only files
    - Makefile.tos: Common makefile for all TOS compilers. If you add further
      ones please email me the new makefile.
    - flex.lnk: Lattice - GST linker argument extension file
    - flex.tlk: Turbo linker argument extension file
     
     
Additional remarks:
     
I didn't add a new initscan.c (= flex -ist scan.l). The current one is good
enough for a first compilation. With this first version of flex you can
rebuild your own scan.c and the final flex version !
     
Minix ST :
  - I had to "chmem =70000 cv" (>50000) to get flex linked
  - 'memset' (PC 1.3, EFTH40,...) is necessary
  - chmem =90000 flex may be sufficient
     
Minix PC :
  It should be possible to port Flex to Minix PC. The current sizes of flex
  are:
                        Minix ST (ACK)    Lattice (TOS)    Turbo (TOS)
     
       size               75300             83305             57957
       compilation time      22'               15'             3'40"
       flex -is scan.l     1'49"               43"               30"
     
  The Minix ST size includes the bad generated code using only a subset of
  the 68000 commands, long addresses only and a huge relocation table.
  Therefore the PC size will be <64 k ! More serious is the fact that I had
  to chmem =90000 flex to get scan.l converted to scan.c . But I never saw
  a more complex lex source than scan.l -- so it should be possible to
  reduce some array sizes without limitation for all day usage.
     
  No one volunteered yet for a Minix PC port -- but if someone will try it
  I would provide him with a new scan.c and some hints.
     
TOS:
  Don't forget to adapt the flexskel path within flexdef.h !
     
     
Bitnet:  V61@DHDURZ1                               Ronald Lamprecht
UUCP:    ...!unido!DHDURZ1.bitnet!V61              Theoretische Physik
ARPAnet: V61%DHDURZ1.BITNET@CUNYVM.CUNY.EDU       (Heidelberg, West Germany)
(Message inbox:36)
Date:          Wed, 05 Jul 89 21:16:07 CET
From:  V61%DHDURZ1.BITNET@csa2.lbl.gov
Subject:       Re: What is TOS
To:  vern@lbl-csam.arpa

TOS is the name of the Atari ST operating system that is very similar
to MSDOS (Both use CR,LF as end of line character). Therefore I have
been astonished that no EOL convertion porblems occur within MSDOS.
     
I have been aware of the double buffering when changing read to fread and  I
accept your argument of a possible slow down. But if you integrate the other
Atari - TOS changes, please insert a
        #ifdef ATARI
              fread ....
        #else
              read  ....
        #endif
in flex.skel .
     
Bitnet:  V61@DHDURZ1                               Ronald Lamprecht
UUCP:    ...!unido!DHDURZ1.bitnet!V61              Theoretische Physik
ARPAnet: V61%DHDURZ1.BITNET@CUNYVM.CUNY.EDU       (Heidelberg, West Germany)




echo x - Makefile_cdiff
sed '/^X/s///' > Makefile_cdiff << '/'
X*** Src_2.1/Makefile	Thu Jun 28 00:06:42 1989
X--- Makefile	Thu Jul  3 02:12:48 1989
X***************
X*** 5,10 ****
X--- 5,11 ----
X  # Porting considerations:
X  #
X  #    For System V Unix machines, add -DSYS_V to CFLAGS.
X+ #    For Minix (ST), add -DSYS_V to CFLAGS
X  #    For Vax/VMS, add -DSYS_V to CFLAGS.
X  #    For MS-DOS, add "-DMS_DOS -DSYS_V" to CFLAGS.  Create \tmp if not present.
X  #         You will also want to rename flex.skel to something with a three
X***************
X*** 21,28 ****
X  SKELETON_DIR = /usr/local/lib
X  SKELETON_FILE = flex.skel
X  SKELFLAGS = -DDEFAULT_SKELETON_FILE=\"$(SKELETON_DIR)/$(SKELETON_FILE)\"
X! CFLAGS = -O
X! LDFLAGS = -s
X  
X  FLEX_FLAGS =
X  FLEX = ./flex
X--- 22,29 ----
X  SKELETON_DIR = /usr/local/lib
X  SKELETON_FILE = flex.skel
X  SKELFLAGS = -DDEFAULT_SKELETON_FILE=\"$(SKELETON_DIR)/$(SKELETON_FILE)\"
X! CFLAGS = -O -DSYS_V
X! LDFLAGS = 
X  
X  FLEX_FLAGS =
X  FLEX = ./flex
X***************
X*** 57,63 ****
X  	yylex.c
X  
X  flex : $(FLEXOBJS)
X! 	$(CC) $(CFLAGS) -o flex $(LDFLAGS) $(FLEXOBJS)
X  
X  first_flex:
X  	cp initscan.c scan.c
X--- 58,65 ----
X  	yylex.c
X  
X  flex : $(FLEXOBJS)
X! 	$(CC) $(CFLAGS) -o flex $(FLEXOBJS) $(LDFLAGS)
X! 	chmem =150000 flex
X  
X  first_flex:
X  	cp initscan.c scan.c
/
echo x - flex.skel_cdif
sed '/^X/s///' > flex.skel_cdif << '/'
X*** Src_2.1/flex.skel	Thu Jun 28 00:19:20 1989
X--- flex.skel	Thu Jul  2 22:18:31 1989
X***************
X*** 28,34 ****
X   * is returned in "result".
X   */
X  #define YY_INPUT(buf,result,max_size) \
X! 	if ( (result = read( fileno(yyin), buf, max_size )) < 0 ) \
X  	    YY_FATAL_ERROR( "read() in flex scanner failed" );
X  #define YY_NULL 0
X  #define yyterminate() return ( YY_NULL )
X--- 28,34 ----
X   * is returned in "result".
X   */
X  #define YY_INPUT(buf,result,max_size) \
X! 	if ( (result = fread( buf,1, max_size, yyin )) < 0 ) \
X  	    YY_FATAL_ERROR( "read() in flex scanner failed" );
X  #define YY_NULL 0
X  #define yyterminate() return ( YY_NULL )
/
echo x - flexdef.h_cdif
sed '/^X/s///' > flexdef.h_cdif << '/'
X*** Src_2.1/flexdef.h	Thu Jun 28 00:43:27 1989
X--- flexdef.h	Thu Jul  3 02:45:50 1989
X***************
X*** 26,31 ****
X--- 26,40 ----
X  
X  /* @(#) $Header: flexdef.h,v 2.0 89/06/20 15:49:50 vern Locked $ (LBL) */
X  
X+ #ifdef ATARI
X+ #define SYS_V
X+ #define abs(x) ((x) < 0 ? -(x) : (x))
X+ #define SHORT_FILE_NAMES
X+ #ifdef TURBO
X+ #define SHORT_EXTERNAL_NAMES
X+ #endif
X+ #endif
X+ 
X  #ifndef FILE
X  #include <stdio.h>
X  #endif
X***************
X*** 41,47 ****
X  #endif
X  
X  #ifndef VMS
X! char *memset();
X  #else
X  /* memset is needed for old versions of the VMS C runtime library */
X  #define memset(s, c, n) \
X--- 50,58 ----
X  #endif
X  
X  #ifndef VMS
X! #ifndef ATARI && TURBO
X! char *memset();
X! #endif
X  #else
X  /* memset is needed for old versions of the VMS C runtime library */
X  #define memset(s, c, n) \
X***************
X*** 81,91 ****
X--- 92,129 ----
X  #define true 1
X  #define false 0
X  
X+ #ifdef ATARI
X+ #define DEFAULT_SKELETON_FILE "D:\\include\\flexskel"
X+ #endif
X+ 
X  
X  #ifndef DEFAULT_SKELETON_FILE
X  #define DEFAULT_SKELETON_FILE "flex.skel"
X  #endif
X  
X+ #ifdef SHORT_EXTERNAL_NAMES
X+ /* avoid long external names that are ambiguous within the first 8 characters */
X+ #define current_mns                 c__mns
X+ #define current_max_rules           c__max_rules
X+ #define current_max_state_type      c__max_state_type
X+ #define current_max_scs             c__max_scs
X+ #define current_max_dfa_size        c__max__size
X+ #define current_max_xpairs          c__max_xpairs
X+ #define current_max_template_xpairs c__max_template_xpairs
X+ #define current_max_dfas            c__max_dfas
X+ #define current_maxccls             c__maxccles
X+ #define current_max_ccl_tbl_size    c__max_ccl_tbl_size
X+ #define indent_puts                 ind_puts
X+ #define indent_put2s                ind_put2s
X+ #define gen_next_compressed_state   gen_n_comressed_state
X+ #define gen_next_match              gen_n_match
X+ #define gen_next_state              gen_n_state
X+ #define variable_trailing_context_rules  var_tr_context_rules
X+ #define variable_trailing_rule           var_tr_rule
X+ #define backtrack_report            backtr_report
X+ #define backtrack_file              backtr_file
X+ #endif
X+ 
X  /* special chk[] values marking the slots taking by end-of-buffer and action
X   * numbers
X   */
X***************
X*** 305,311 ****
X      int int_val;
X      } ;
X  
X! typedef struct hash_entry *hash_table[];
X  
X  #define NAME_TABLE_HASH_SIZE 101
X  #define START_COND_HASH_SIZE 101
X--- 343,349 ----
X      int int_val;
X      } ;
X  
X! typedef struct hash_entry **hash_table;
X  
X  #define NAME_TABLE_HASH_SIZE 101
X  #define START_COND_HASH_SIZE 101
X***************
X*** 372,378 ****
X  extern int datapos, dataline, linenum;
X  extern FILE *skelfile, *yyin, *temp_action_file, *backtrack_file;
X  extern char *infilename;
X! extern char action_file_name[];
X  
X  
X  /* variables for stack of states having only one out-transition:
X--- 410,416 ----
X  extern int datapos, dataline, linenum;
X  extern FILE *skelfile, *yyin, *temp_action_file, *backtrack_file;
X  extern char *infilename;
X! extern char *action_file_name;
X  
X  
X  /* variables for stack of states having only one out-transition:
/
echo x - main.c_cdiff
sed '/^X/s///' > main.c_cdiff << '/'
X*** Src_2.1/main.c	Thu Jun 28 00:30:39 1989
X--- main.c	Thu Jul  3 02:27:47 1989
X***************
X*** 81,96 ****
X  FILE *temp_action_file;
X  FILE *backtrack_file;
X  int end_of_buffer_state;
X! #ifndef SHORT_FILE_NAMES
X! char action_file_name[] = "/tmp/flexXXXXXX";
X! #else
X! char action_file_name[] = "flexXXXXXX.tmp";
X! #endif
X! 
X  #ifndef SHORT_FILE_NAMES
X  static char outfile[] = "lex.yy.c";
X  #else
X! static char outfile[] = "lexyy.c";
X  #endif
X  static int outfile_created = 0;
X  
X--- 81,91 ----
X  FILE *temp_action_file;
X  FILE *backtrack_file;
X  int end_of_buffer_state;
X! char *action_file_name;
X  #ifndef SHORT_FILE_NAMES
X  static char outfile[] = "lex.yy.c";
X  #else
X! static char outfile[] = "lex_yy.c";
X  #endif
X  static int outfile_created = 0;
X  
X***************
X*** 328,333 ****
X--- 323,329 ----
X      {
X      int i, sawcmpflag, use_stdout;
X      char *arg, *skelname = NULL, *flex_gettime(), clower(), *mktemp();
X+     char *tmp_action =(char *)0, *malloc();
X  
X      printstats = syntaxerror = trace = spprdflt = interactive = caseins = false;
X      backtrack_report = performance_report = ddebug = fulltbl = fullspd = false;
X***************
X*** 349,354 ****
X--- 345,355 ----
X  	for ( i = 1; arg[i] != '\0'; ++i )
X  	    switch ( arg[i] )
X  		{
X+ 		case 'a':
X+ 		    if ( i != 1 )
X+ 			flexerror( "-a flag must be given separately" );
X+ 		    tmp_action = &arg[i+1];
X+ 		    goto get_next_arg;
X  		case 'b':
X  		    backtrack_report = true;
X  		    break;
X***************
X*** 445,452 ****
X  		    printstats = true;
X  		    break;
X  
X! 		default:
X! 		    lerrif( "unknown flag %c", (int) arg[i] );
X  		    break;
X  		}
X  
X--- 446,458 ----
X  		    printstats = true;
X  		    break;
X  
X! 		case '?':
X! 		    flexinfo(0);
X! 		    break;
X! 
X! 		default:
X! 		    fprintf(stderr,"flex : unknown flag %c\n", (int) arg[i] );
X! 		    flexinfo(1);
X  		    break;
X  		}
X  
X***************
X*** 454,459 ****
X--- 460,493 ----
X  	;
X  	}
X  
X+ 
X+ /* if you change the default tmp file names don't forget to change the
X+    initialization for i, too !
X+    
X+ */
X+ #ifndef SHORT_FILE_NAMES
X+     i = 10;
X+     if (!tmp_action) i += 5;
X+ #else
X+     i = 12;
X+ #endif
X+     if (tmp_action) 
X+         i += strlen(tmp_action) + 1;
X+     if((action_file_name = malloc(i+1)) == (char *)0)
X+         flexerror("No memory for action_file_name");
X+     *action_file_name = (char) NULL;
X+     if (tmp_action) 
X+         strcat(action_file_name,tmp_action);
X+ #ifndef SHORT_FILE_NAMES
X+     else
X+         strcat(action_file_name,"/tmp");
X+     strcat(action_file_name,"/flexXXXXXX");
X+ #else
X+     if (tmp_action)
X+         strcat(action_file_name,"/");
X+     strcat(action_file_name,"flexXXXX.tmp");
X+ #endif
X+     
X      if ( (fulltbl || fullspd) && usemecs )
X  	flexerror( "full table and -cm don't make sense together" );
X