build: work around GCC warnings on Flex code

With GCC 7.3.0 and Flex 2.6.4, we get warnings on all the generated
scanners:

    examples/calc++/calc++-scanner.cc: In function 'void yyrestart(FILE*)':
    examples/calc++/calc++-scanner.cc:1611:20: error: potential null pointer dereference [-Werror=null-dereference]
     /* %endif */
      ~~~~~~~~~~~       ^
    examples/calc++/calc++-scanner.cc:1607:19: error: potential null pointer dereference [-Werror=null-dereference]
     /* %if-c-only */
      ~~~~~~~~~~~~~~~  ^
    examples/calc++/calc++-scanner.cc:1611:20: error: potential null pointer dereference [-Werror=null-dereference]
     /* %endif */
      ~~~~~~~~~~~       ^
    examples/calc++/calc++-scanner.cc:1607:19: error: potential null pointer dereference [-Werror=null-dereference]
     /* %if-c-only */
      ~~~~~~~~~~~~~~~  ^
    cc1plus: all warnings being treated as errors

Obviously the lines are incorrect, and the warnings are emitted twice.
Still, let's get rid of these warnings.

* doc/bison.texi, src/flex-scanner.h: Disable these warnings in code
generated by Flex.
This commit is contained in:
Akim Demaille
2018-08-15 13:49:01 +02:00
parent 7783ba2d4f
commit ef98967ada
2 changed files with 20 additions and 0 deletions

View File

@@ -44,6 +44,20 @@ int FLEX_PREFIX (lex_destroy) (void);
#define last_string FLEX_PREFIX (last_string)
// Pacify warnings in yy_init_buffer (observed with Flex 2.6.4 and GCC
// 7.3.0).
//
// ./src/scan-skel.c: In function 'skel_restart':
// ./src/scan-skel.c:2035:20: error: potential null pointer dereference [-Werror=null-dereference]
// b->yy_fill_buffer = 1;
// ~~~~~~~~~~~~~~~~~~^~~
// ./src/scan-skel.c:2031:19: error: potential null pointer dereference [-Werror=null-dereference]
// b->yy_input_file = file;
// ~~~~~~~~~~~~~~~~~^~~~~~
#if defined __GNUC__ && 7 <= __GNUC__
# pragma GCC diagnostic ignored "-Wnull-dereference"
#endif
/* It seems to be a nice "feature" of Flex that one cannot use yytext,
yyleng etc. when a prefix is given, since there is no longer a
#define, but rather the token is actually changed in the output.