Re: Wide STRING_CSTs

[Geoff Keating <geoffk at geoffk dot org>]

"Joseph S. Myers" <jsm28@cam.ac.uk> writes:

> What is the (undocumented) format of a wide string STRING_CST meant to be?  
> lex_string seems to treat it as being stored as an array of target wchar_t
> (in the target endianness), but combine_strings seems to treat it as being
> stored in the host endianness.  Which of these is right?  I'll document
> the conclusion for c-tree.texi.

I would think it should be target endianness, because the size of each
character is the size of the target's wchar_t.

> As is, it seems (from the source and examining assembler output) that the
> following testcase will fail for cross-compiles to architectures of
> different endianness.  OK to commit it to gcc.c-torture/execute/ on
> mainline and branch?  (It's an execution test rather than a link-time test
> since GCC doesn't seem to optimise accesses to wide string literals here.)

The test is OK.  Please include a ChangeLog entry when you commit.

> /* Origin: Joseph Myers <jsm28@cam.ac.uk>.
> 
>    This tests for inconsistency in whether wide STRING_CSTs use the host
>    or the target endianness.  */
> 
> extern void exit (int);
> extern void abort (void);
> 
> int
> main (void)
> {
>   if (L"a" "b"[1] != L'b')
>     abort ();
>   exit (0);
> }

-- 
- Geoffrey Keating <geoffk@geoffk.org>