[9fans] new compiler...

[9fans] new compiler...

[Steve Simon]

Hi,

I am trying to push some old code through the new
compilers which are reporting loads of "incompatible
type signatures" errors.

When I look at the code I cannot see what the problem is.

Could somone summerise in a sentence or two what the 
increase in sensitivity to type errors is?

BTW the dynld(2) manpage referenced in ld(2) seems not
to have made it out.

Thanks

-Steve

Re: [9fans] new compiler...

[Charles Forsyth]

[-- Attachment #1: Type: text/plain, Size: 1144 bytes --]

it might be caused by a genuine error (usually an extern declared differently in two .c files
rather than in a common .h), or often by a construction similar to the following:

	file1.h:
		typedef struct Secret Secret;

		void fred(Secret*);

	file1.c
		#include "file1.h"

		/* use fred */

	file2.c
		#include "file1.h"

		struct Secret {
			...
		};

		void
		fred(Secret *s)
		{
			...
		}

so that in scope file1.c the signature of Secret (and thus fred) is different
from that in file2.c.  follow the declaration of Secret in file1.h by

	#pragma incomplete Secret

to ensure that although file2.c declares Secret in full,
the full declaration is ignored when calculating
the  signature of Secret and fred, producing the same result
for signof in file2 as in file1.c

(there are examples in the kernel sources, see portdat.h)

i added a way for the #pragma to switch off the checking for
basket cases such as freetype, but that change seems to have gone missing,
even from my own copies.
i'll find it.

you can comment out the -T option in CFLAGS in /sys/src/mkfile.proto
if you're in a hurry

[-- Attachment #2: Type: message/rfc822, Size: 2916 bytes --]

From: "Steve Simon" <steve@quintile.net>
To: 9fans@cse.psu.edu
Subject: [9fans] new compiler...
Date: Fri, 7 Jan 2005 17:36:31 0000
Message-ID: <f9021c0a5d1a1e02379559b027768742@quintile.net>

Hi,

I am trying to push some old code through the new
compilers which are reporting loads of "incompatible
type signatures" errors.

When I look at the code I cannot see what the problem is.

Could somone summerise in a sentence or two what the 
increase in sensitivity to type errors is?

BTW the dynld(2) manpage referenced in ld(2) seems not
to have made it out.

Thanks

-Steve

Re: [9fans] new compiler...

[jmk]

That's my fault, the setting of the -T option to CFLAGS in /sys/src/mkfile.proto
was not supposed to escape to sources yet. Take it out if it causes problems.

--jim

Re: [9fans] new compiler...

[Charles Forsyth]

>>i added a way for the #pragma to switch off the checking for
>>basket cases such as freetype, but that change seems to have gone missing,
>>even from my own copies.

i've found it in the dump, so i'll submit a patch shortly.