Re: [PATCH] Libiberty for VMS - mkstemps.c don't mix case

[Douglas B Rupp <rupp@gnat.com>]

DJ Delorie wrote:
>> VMS has file versioning, I think it would detect it only if the
>> version was specified or the directory was created with single
>> versioning.
> 
> Doh!  I forgot about that.
> 
> Ok, that leads to a new question - shouldn't we specify a version, or
> single versioning, when creating temp files, so that someone else
> can't just create a new copy of the same file with a newer version and
> trip up the compiler?  Or do the file permissions pass on to new
> versions?
> 

At long last getting back to this discussion.  I did some experiments 
with forcing the temp file to have a version number for the .s temp file 
created by gcc and passed to cc1.  Creating the temp file works fine, 
but cc1 now gets the fatal error:
"can't open <sometempdir>/cccadaaa.s.1 for writing: file exists"

The problem comes from toplev.c/init_asm_output
asm_out_file = fopen (asm_file_name, "w+b");

The "w+" attribute says open a new file for writing. The file exists, 
the filename has a version number on it already, so VMS generates this 
error.

If I change the attributes to "a+b" it works fine on VMS.

So what now?  Seems to me the VMS behavior is entirely rational in 
insisting on a new version and that the Unix behavior is somewhat 
arbitrary in ignoring the fact the file already exists.

What is your thinking?