Re: AC_PROG_LEX broken on IRIX with MIPSpro Compilers

[Peter O'Gorman]

Paul Eggert wrote:
> "Peter O'Gorman" <address@hidden> writes:
> 
>> -e { yyless (input () != 0); }
>> +e { yyless ((input () != 0)); }
> 
> Why is this patch necessary?  yyless is supposed to take an int
> parameter, and another set of parentheses shouldn't affect the
> parameter's type or value.

After macro expansion that line turns into:
{ do { *yy_cp = yy_hold_char;  yy_c_buf_p = yy_cp = yy_bp + input () !=
0 - 0; yytext = yy_bp; yyleng = (int) (yy_cp - yy_bp); yy_hold_char =
*yy_cp; *yy_cp = '\0'; yy_c_buf_p = yy_cp;; } while ( 0 ); }

The compiler barfs an error with:
yy_c_buf_p = yy_cp = yy_bp + input () != 0 - 0;

Reducing to:
yy_cp = yy_bp + input () != 0;

The compiler errors out with:
  A value of type "int" cannot be assigned to an entity of type "char *".

  yy_cp = yy_bp + input () != 0;
        ^

1 error detected in the compilation of "cpp.c"


Now, I am not sure if the compiler is right or wrong, and am not really
bothered either way, the AC_PROG_LEX test is producing incorrect results
on irix6.5 because of this error.

Now, looking at the expanded version, I am unsure if my patch is
correct, is the test supposed to be that input() != 0 or that (yy_bp +
input ()) != 0 ? At the moment, due to operator precedence, it is
testing that  yy_bp + input () is non-zero, is that intentional?

Peter
-- 
Peter O'Gorman
http://pogma.com