notes.txt

C++版 词法分析、语法分析器

========================================================
Date: 21 Jul 93 14:30:10-0400
From: John Holmes Sr <[e-mail]>
Subject: Re: flex++/bison++
To: [e-mail]

I tried all possible alternatives but I think there is still a problem
with the main.c file in your distribution of flex++. The algorithm for 
generating the variable "headerfilename" actually extracts the base name of 
the file supplied in the command line (and NOT the name of the scanner as 
defined by the %name in flex++) and then uses this extracted string to
derive the #ifdef name for the flex header anti-include mechanism.
However, what happens there is that any number that appears in a filename
is converted to the symbol '_'.  Everything will be fine if the file names
do not end with a number. But if they do, then multiple file names with
the same prefix and different only in the trailing numbers get mapped onto
the same #ifdef guard.   To fix this bug (if it is a bug) I went ahead and
modified the  main.c file to take care of digits appearing in the file
name.  Now it works fine.



From coetmeur Thu Jul 22 09:21:23 1993
To: [e-mail]
Subject: Re: flex++/bison++



	THANKS, and good job ! I am patching the sources in this
direction.  
 
========================================================

From coetmeur Mon Jul 19 13:07:23 1993
To: [e-mail]
Subject: Re: flex++
Content-Length: 2153
X-Lines: 53
Status: RO

	Good job ! you find one ! With such help, things are going
	better and better.

	Effectively, I have forgotten the #define YY_USE_CLASS in the
    flexskel.cc in the #ifdef __cplusplus ... It was not evident,
    because the #include of the header contain such a #define , and it
    is included short time after. But the effect if to keep the old
    behaviour of YYINPUT and so on... so no virtual were used.  I don't
    understand why I didn't found it before... but you find it, and so
    I send you  may best thanks. 

	To patch it, just add this '#define YY_USE_CLASS' between the
    #ifdef __cplusplus and the #endif, around line 39. I have patched
    it in my master copy (release 2.3.7-4), in a similar way (I just
    added an anti-redefine #ifndef ). I hope it doesn't add bugs !!

	I tested this on my own tools, and it does work. I noted a
    change in the behaviour of the scanner. Since we now really use
    buffered IO and not read, the scanner does buffers much data and so
    output is not synchronized to input on keyboard/screen IO. this
    shows that it works ! (hope so)



	Thanks for help... Hope my work will be useful...

	Alain Coetmeur

PS: You should rename the install directory of flex++ and
bison++ to something like bison++-1.19-2 and flex++-2.3.8-3 to avoid
mismatch with official flex and bison. I confirm that my job is not
official, and not at all the successor of flex and bison. It is just
a cleanly patched version, kept as much compatible as possible, in
my mind.

========================================================

From coetmeur Mon Jul 12 18:51:02 1993
To: [e-mail]
Subject: Re: bison++/flex++ / test bigtest
Content-Length: 1919
X-Lines: 47
Status: RO



> The distribution of flex++
> does not properly resolve a "make test" or "make bigtest" command.  I
> get lots of errors.  Is this anything I need to worry about?


	Good remark. there was 3 problems  .

		1- I have not regenerated the 'initscan.c' (or scan.c)
		 it is not important since it works, but is not coherent.
		If you want to be clean :

		touch scan.l ; make flex++ initscan 

		2- The new system insert #line directive that use the skeletons name,
		and since the test skeleton is local (./flexskel.cc .h) and
		the other global (/u/.../flexskel.cc .h) there is always differences.

		3- the new code use trailing context in the internal scanner (scan.l)
		and so we can no more use fast table compression mode to generate
		scan.c itself. so don't worry errors are logical.

here is the new targets for test ...
but you should not bother with test and bigtest , it is obsolete !!


test : flex++
	echo "diff should only be on #line directive"
	-./flex++ $(FLEX_FLAGS) -S./flexskel.cc -H./flexskel.h $(COMPRESSION) scan.l | diff scan.c -
bigtest :
	echo "diff should only be on #line directive"
	-rm -f scan.c ; $(MAKE) COMPRESSION="-C" test
	echo "diff should only be on #line directive"
	-rm -f scan.c ; $(MAKE) COMPRESSION="-Ce" test
	echo "diff should only be on #line directive"
	-rm -f scan.c ; $(MAKE) COMPRESSION="-Cm" test
	echo this should fail because there is trailing context in the grammar
	-rm -f scan.c ; $(MAKE) COMPRESSION="-Cfe" test
	echo this should fail because there is trailing context in the grammar
	-rm -f scan.c ; $(MAKE) COMPRESSION="-CFe" test
	echo this should fail because there is trailing context in the grammar
	-rm -f scan.c ; $(MAKE) COMPRESSION="-Cf" test
	echo this should fail because there is trailing context in the grammar
	-rm -f scan.c ; $(MAKE) COMPRESSION="-CF" test
	echo "diff should only be on #line directive"
	-rm -f scan.c ; $(MAKE)

=====================================================
> Date: 29 Aug 93 15:13:11+0200
> From: Andreas Holzmann <[e-mail]>
> Subject: flex++: header in section2?
> To: [e-mail]
> 
> I have some problems with your flex++ version V3.0.2, based on 2.3.8.
> In line 301 of flexskel.h there is an 'announcement' for the section2
> %header{ but it was impossible to me to get some lines at this point of
> the header file.
	Ok, it is a bug
> 
> I tried the following program:
> (translated with: flex++ -otest.cc -htest.h test.l)
> -------- start test.l ------
> %name test
> %header{
> #define HEADER_TEST_1 1
> %}
> 
> %%
> 
> %header{
> #define HEADER_TEST_2 1
> %}
> 
> %%
> -------- end test.l ------
> 
> the output of test.h after line 55:
> 
> -------- start test.h ------
> /* % here is the declaration from section1 %header{  */
> #line 6 "test.l"
> #define HEADER_TEST_1 1
> #line 10 "test.l"
> #line 12 "test.l"
> #define HEADER_TEST_2 1
> #line 16 "test.l"
> #endif
> #line 52 "./flexskel.h"
> ------ end test.h ------
> 
> well, I expected the HEADER_TEST_2 line to be at the end of the test.h file.

	Right, exactly as you say ! there is a bug in scan.l.
	I must load the second skeleton  section (which is empty, this is why
        I didn't notice it)

	i have corrected it. I just added the header_skeleton_out(); in the
       lex rule to pass from section 1 to 2



^"%%".*			{
			set_lexer_name((char *)0);
			sectnum = 2;
			line_directive_out( headerfile );
      			if(headerfilename!=NULL)
		        {
      			 fprintf(stdout,
                          "#include \"%s\"\n",includefilename);
 >>>			 header_skeleton_out();
    		 	};
			line_directive_out( stdout );
			BEGIN(SECT2PROLOG);
			return ( SECTEND );
			}
> 
> Something strange happend, when I delete '%name test' from test.l
> Then the HEADER_TEST_1 line moves to the top of test.h

	when you don't give the %name, the flex++ know thereof the 
   name of the parser, so it decide immediately to include the parser declaration.
   but before this, it cannot, since the name of the parser is needed to 
   generate the symbols. when you don't give any %name, the
   %% simulate a %name parser, just before the %%, up to be coherent, and
   so it load the class declaration before section 2.


> whereas the
> HEADER_TEST_2 line furthermore is at the (wrong?) place.
 	now it will be a the good place
> 
> so my question is, how do I get some lines after this line:
> /* % here is the declaration from section2 %header{ */
	now it will work

thanks for all