Re: fast-fixinc losing badly on alpha-dec-osf4.0a

Re: fast-fixinc losing badly on alpha-dec-osf4.0a

[Jeffrey A Law]

  In message < 19990603142433.21538.qmail@egcs.cygnus.com >you write:
  > > They fall into two basic categories.
  > > 
  > >   1. Failure to fix the _IO macros in many files.
  > 
  > I will work on this today:
  > 
  > Applying No_Double_Slash          to ./bktest/fix.h  
  > Applying Io_Def_Quotes            to ./bktest/fix.h
  > Fixed:  ./bktest/fix.h
  > 
  > Io_Def_Quotes did not do anything :-(
Looks like your fix did the trick on these.  Thanks!

  > >   2. Failure to turn C++ comments into a C comment.
  > 
  > Very odd.  What is the name of the file?  My fixincl program
  > stripped out the '//' comment from your sample.
  > 
  > The no_double_slash fix omits all files containing any of
  > 
  >     CC/
  >     cxx/
  >     ++/
  > 
  > in their names.  Also, the fix is only applied if either:
OK.  One of them has a ++.  It appears inside a macro like this:

#define PROBE_START(PD,a,b,c,d,e)                       \
MACRO_BEGIN                                             \
if (probe_smp) {                                        \
   probe_config_t *PT = (PD)->pb_table;                 \
   int np;                                              \
   np = (PT) ? PT->type : 0;                            \
   lock_read(&(PD)->pb_lock);                           \
   if (np > 1) {                                        \
      unsigned int flag;                                \
      int j;                                            \
      for (flag=0, j=0; j<np; j++)                      \
        if (lock_islocked((PT)->probe[j]))              \
          flag |= (unsigned int)1<<(np-1-j);            \
      PT->count[flag]++;                                \
   }                                                    \
}                                                       \
MACRO_END

In another one of the files CC appears like this:

#define K2_MSTR_INTR_MODE_ACCEL 1       /* Accelerator Mode     */

I suspect we need to be a little more selective about when the fix does
not apply.  What a mess :(

  >  *  traditional hacks like irix_bogus_cxx_cmnt, no longer work (which
  >  *  strangely enough was also used on alpha-dec-osf4.0d).  If we skip the
  >  *  hack whenever we see ``"//"'', then the need for the secondary hack
  >  *  disappears.  Note: it is painful to ensure that the first quote
  >  *  exists, so we just check for the trailing quote directly abutting
  >  *  the //.  Note: We should never touch a line that has // completely

Note the reference to osf4.0d.  I wonder if it's related to the problem
we're seeing.


  > BTW, what is your intent vis-a-vis
  > "fixinc patch:  eliminate need for setenv/putenv"?
I installed it last night.

I'm going to go ahead and install the fixes you sent today :-)

Note, we're also not fixing sys/limits.h, but I already know what the problem
is with that one and will fix it shortly.

jeff

RE: fast-fixinc losing badly on alpha-dec-osf4.0a

[korbb]

Reply-To: ddsinc09@ix.netcom.com
*PLEASE* if you press the "Reply" button, edit my address.  Thanks.


Hi Jeff,

> fast-fixincludes is missing several fixes that were performed
> by the old fixincludes script on alpha-dec-osf4.0a.
> 
> They fall into two basic categories.
> 
>   1. Failure to fix the _IO macros in many files.

I will work on this today:

Applying No_Double_Slash          to ./bktest/fix.h  
Applying Io_Def_Quotes            to ./bktest/fix.h
Fixed:  ./bktest/fix.h

Io_Def_Quotes did not do anything :-(

>   2. Failure to turn C++ comments into a C comment.

Very odd.  What is the name of the file?  My fixincl program
stripped out the '//' comment from your sample.

The no_double_slash fix omits all files containing any of

    CC/
    cxx/
    ++/

in their names.  Also, the fix is only applied if either:

    1.  the double slash does not appear to be part of a URL
    2.  is not followed by a double quote or an asterisk

/*
 *  Remove the double-slash comments
 *  They *must* be removed so it will not create nested comments!!
 *  However, they will *not* be removed if the file is from a C++ dir.
 *
 *  There *used* to be a number of similar problems in various OSes:

 *  Turning // comments into normal comments trashes this IRIX 4.0.1
 *  header file, which embeds // comments inside multi-line
 *  comments.  If this looks like the IRIX header file, we refix it by
 *  just throwing away the // comments.

 *  Same problem with a file from SunOS 4.1.3 : a header file containing
 *  the string "//" embedded in "/ * * /"

 *  There is a similar problem with the VxWorks drv/netif/if_med.h file.

 *  And also with the HP-UX 10 and HP-UX 11 sys/pci.h file

 *  Now that we delete the // comments instead of converting them to / * * /,
 *  traditional hacks like irix_bogus_cxx_cmnt, no longer work (which
 *  strangely enough was also used on alpha-dec-osf4.0d).  If we skip the
 *  hack whenever we see ``"//"'', then the need for the secondary hack
 *  disappears.  Note: it is painful to ensure that the first quote
 *  exists, so we just check for the trailing quote directly abutting
 *  the //.  Note: We should never touch a line that has // completely
 *  within quotes but this is somewhat hard to check for.

 *  Ultimately, this fix ought to go inside of C code where
 *  we can do a better analysis on the need and method for fixing.
 */
fix = {
    hackname = no_double_slash;
    /*
     *  Test that the file-to-fix does not from a C++ directory
     */
    test   = '-z `echo ${file} | egrep \'(CC|cxx|\+\+)/\' `';

    /*
     *  Select only files that contain a double slash that either:
     *  1.  starts a line
     *  2.  does not follow a colon (not part of a URL)
     *  3.  is not immediately followed by '"' or '*'
     */
    select = '(^|[^:])//[^"*]';

    /*
     *  Remove the comments that start a line
     */
    sed    = 's,^//.*$,,';

    /*
     *  Remove the comments that are not quoted and not in a URL
     */
    sed    = 's,[^:]//[^"].*$,,';
};

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


BTW, what is your intent vis-a-vis
"fixinc patch:  eliminate need for setenv/putenv"?
Thanks.
        Bruce

fast-fixinc losing badly on alpha-dec-osf4.0a

[Jeffrey A Law]

fast-fixincludes is missing several fixes that were performed by the old
fixincludes script on alpha-dec-osf4.0a.

They fall into two basic categories.

  1. Failure to fix the _IO macros in many files.

  2. Failure to turn C++ comments into a C comment.


This was with the script version of fast-fixinc.  Unfortunately, the files
are copyrighted by digital.  I will include a a couple examples to show what
we are not properly handling.  Hopefully they will be enough to find the
generic problem and fix all the occurences (there's at least a dozen files
which are not properly fixed).

In the first group we need to be quoting the 'a'.  In the second group we
need to fix up the c++ comment in the first line.

/* ioctl commands */
#define IO_BIO		_IOW(a,1,0)	/* Block I/O commands */
#define IO_SYNC		_IO(a,2)	/* Synchronize host with AP */
#define IO_WR		_IO(a,3)	/* Synchronize AP Running with host */
#define IO_WD		_IO(a,4)	/* Synchronize AP DMA with host */
#define IO_FPSERR	_IOR(a,5,short)	/* return EFPS error information */
#define IO_RCSR		_IOR(a,6,short)	/* read CSR register */
#define IO_RFN		_IOR(a,7,short)	/* read FN register */
#define IO_SETMODE	_IOW(a,8,short)	/* sets low byte of drive mode word */
#define IO_GETMODE	_IOR(a,9,short)	/* gets driver mode word */
#define IO_RESET	_IO(a,10)	/* reset the AP */


//#define	EBM10_MIN_SLOT		5	/* PCI_AD<16> */
/*#define	EBM10_MIN_SLOT		7 */	/* PCI_AD<18> */