cpplib and endianness problems

cpplib and endianness problems

[Neil Booth]

I noticed a few weeks ago that Sparc was failing on gcc.dg/cpp/if-2.c

Unfortunately, there was more than one test on the line it was failing
on, so I separated the tests to see which one it was.

From recent posts to gcc-testresults, like Kaveh's, we see the
following fails on Sparc:

#if L'\xfeed' != 0xfeed
#error 0xfeed	/* { dg-bogus "error" "wide charconst recognition 2" } */
#endif

I have no idea if this test is valid; i.e. how should we be
interpreting L'\xfeed'?  Should it be equal to 0xfeed?

Neil.

Re: cpplib and endianness problems

[Dave Brolley]

Neil Booth wrote:

> I noticed a few weeks ago that Sparc was failing on gcc.dg/cpp/if-2.c
> 
> Unfortunately, there was more than one test on the line it was failing
> on, so I separated the tests to see which one it was.
> 
> >From recent posts to gcc-testresults, like Kaveh's, we see the
> following fails on Sparc:
> 
> #if L'\xfeed' != 0xfeed
> #error 0xfeed	/* { dg-bogus "error" "wide charconst recognition 2" } */
> #endif
> 
> I have no idea if this test is valid; i.e. how should we be
> interpreting L'\xfeed'?  Should it be equal to 0xfeed?

The escape sequence, \xfeed represents one character of type wchar_t. 
This character is then converted to type 'int'. So the answer depends on 
the size of wchat_t and whether it is signed or not.

Dave

Re: cpplib and endianness problems

[Neil Booth]

Dave Brolley wrote:-

> > #if L'\xfeed' != 0xfeed
> > #error 0xfeed	/* { dg-bogus "error" "wide charconst recognition 2" } */
> > #endif
> 
> The escape sequence, \xfeed represents one character of type wchar_t. 
> This character is then converted to type 'int'. So the answer depends on 
> the size of wchat_t and whether it is signed or not.

Hi Dave,

Sorry to take so long to get back to you on this.

wchar_t should be >= 16 bits everywhere, right?  Similarly for int.
Certainly on Solaris.  I don't see why the test fails.

Is there a generally correct form for the above test?  I'm a little
confused.

Neil.

Re: cpplib and endianness problems

[Jamie Lokier]

Neil Booth wrote:
> > > #if L'\xfeed' != 0xfeed
> > > #error 0xfeed	/* { dg-bogus "error" "wide charconst recognition 2" } */
> > > #endif
> > 
> > The escape sequence, \xfeed represents one character of type wchar_t. 
> > This character is then converted to type 'int'. So the answer depends on 
> > the size of wchat_t and whether it is signed or not.
> 
> Hi Dave,
> 
> Sorry to take so long to get back to you on this.
> 
> wchar_t should be >= 16 bits everywhere, right?  Similarly for int.
> Certainly on Solaris.  I don't see why the test fails.
> 
> Is there a generally correct form for the above test?  I'm a little
> confused.

Does it pass with 0x1234?  If so, wchar_t is signed and getting
sign-extended.

-- Jamie

Re: cpplib and endianness problems

[Neil Booth]

Jamie Lokier wrote:-

> Does it pass with 0x1234?  If so, wchar_t is signed and getting
> sign-extended.

Ah, I see.  OK, this adds that test too; let's see.  If this passes,
I'll remove the other one.

Neil.

	* gcc.dg/cpp/if-2.c: Add wide char test without sign extension.

Index: gcc.dg/cpp/if-2.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/testsuite/gcc.dg/cpp/if-2.c,v
retrieving revision 1.2
diff -u -p -r1.2 if-2.c
--- if-2.c	2000/12/11 07:46:25	1.2
+++ if-2.c	2001/01/11 07:51:54
@@ -13,6 +13,10 @@
 #error 0xfeed	/* { dg-bogus "error" "wide charconst recognition 2" } */
 #endif
 
+#if L'\x1234' != 0x1234
+#error 0xfeed	/* { dg-bogus "error" "wide charconst recognition 2" } */
+#endif
+
 #if 'abcd' /* { dg-warning "multi-character character constant" "multi-character charconst" } */
 #endif
 

Re: cpplib and endianness problems

[Fergus Henderson]

On 11-Jan-2001, Neil Booth <neil@daikokuya.demon.co.uk> wrote:
> +#if L'\x1234' != 0x1234
> +#error 0xfeed	/* { dg-bogus "error" "wide charconst recognition 2" } */

Don't you want `#error 0x1234' there?

-- 
Fergus Henderson <fjh@cs.mu.oz.au>  |  "I have always known that the pursuit
                                    |  of excellence is a lethal habit"
WWW: < http://www.cs.mu.oz.au/~fjh >  |     -- the last words of T. S. Garp.

Re: cpplib and endianness problems

[Neil Booth]

Fergus Henderson wrote:-

> Don't you want `#error 0x1234' there?

Yes, just did that :-)

Neil.

Re: cpplib and endianness problems

[Jamie Lokier]

Neil Booth wrote:
> > Does it pass with 0x1234?  If so, wchar_t is signed and getting
> > sign-extended.
> 
> Ah, I see.  OK, this adds that test too; let's see.  If this passes,
> I'll remove the other one.

I wonder if it makes Unicode sense to have a 16 bit signed wchar_t?

-- Jamie

Re: cpplib and endianness problems

[Neil Booth]

Jamie Lokier wrote:-

> Does it pass with 0x1234?  If so, wchar_t is signed and getting
> sign-extended.

Yes, it appears OK.  So I'll comment out the old test, with this patch.

Neil.

	* gcc.dg/cpp/if-2.c: Comment out occasionally bogus test; we
	have an equivalent working one below it.

Index: gcc.dg/cpp/if-2.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/testsuite/gcc.dg/cpp/if-2.c,v
retrieving revision 1.4
diff -u -p -r1.4 if-2.c
--- if-2.c	2001/01/11 08:11:29	1.4
+++ if-2.c	2001/01/15 07:47:12
@@ -9,8 +9,10 @@
 #error L'a'	/* { dg-bogus "error" "wide charconst recognition 1" } */
 #endif
 
+#if 0 /* This test doesn't work on targets with signed 16-bit wchar_t.  */
 #if L'\xfeed' != 0xfeed
 #error 0xfeed	/* { dg-bogus "error" "wide charconst recognition 2" } */
+#endif
 #endif
 
 #if L'\x1234' != 0x1234

Re: cpplib and endianness problems

[Alexandre Oliva]

On Jan 15, 2001, Neil Booth <neil@daikokuya.demon.co.uk> wrote:

> Jamie Lokier wrote:-
>> Does it pass with 0x1234?  If so, wchar_t is signed and getting
>> sign-extended.

> Yes, it appears OK.  So I'll comment out the old test, with this patch.

I think it would be nice to check whether we do The Right Thing (TM)
when wchar_t is signed too.  How about this patch?  Ok to install?

Re: cpplib and endianness problems

[Neil Booth]

Alexandre Oliva wrote:-

> I think it would be nice to check whether we do The Right Thing (TM)
> when wchar_t is signed too.  How about this patch?  Ok to install?

Yes, great thanks.

Neil.