Re: Disallow checking for upper case characters in gfc_match

[Tobias Burnus <burnus at net-b dot de>]

Daniel Kraft wrote:
> while working on parsing of typebound-procedures I just spent some 
> time puzzling over why
>   gfc_match (" PASS")
>
> What do you think about the attached patch to add an additional 
> "sanity check" to gfc_match resulting in an ICE if a literal character 
> is upper-case?  This is very simple, has not much potential to break 
> anything and would have saved me some time if it were already there.
I think it is ok to add a check, however, as match is quite often called 
and the chance is quite low that this happens:

How about using gcc_assert instead? The result is the same (except that 
the error message is not as clear) and it is not checked when 
ENABLE_RUNTIME_CHECKING is disabled. I think it is unlikely to occur in 
user code, where a better error message is helpful.

> Ok to commit with an appropriate ChangeLog entry?  Regression testing 
> running on GNU/Linux-x86-32.
Thus: OK to add a gcc_assert.

Tobias

PS: Please CC patches to gcc-patches@.
PPS:  s/upper case/lower case not/ or s/Expecting/Unexpected/ in the 
gfc_internal_error. But this is not needed for gcc_assert ;-)

Index: match.c
===================================================================
--- match.c	(revision 138086)
+++ match.c	(working copy)
@@ -1187,6 +1187,12 @@ loop:
	}

    default:
+
+      /* gfc_next_ascii_char converts characters to lower-case, so we shouldn't
+	 expect an upper case character here!  */
+      if (TOLOWER (c) != c)
+	gfc_internal_error ("gfc_match(): Expecting upper case '%c'", c);
+
      if (c == gfc_next_ascii_char ())
	goto loop;
      break;