RE: [PATCH] PE direct linking to dlls, accept any filename.

[Danny Smith <dannysmith at clear dot net dot nz>]

Pedro Alves
Tuesday, 19 December 2006 12:47 a.m.
> To: Danny Smith
> Cc: 'Binutils'
> Subject: Re: [PATCH] PE direct linking to dlls, accept any filename.
> 
> 
> Danny Smith wrote:
> 
> > Could you add a test for symlinks (on cygwin), eg where 
> libfoo.a is a
> > symlink to $PATH/foo.dll.
> > This is what 
> >        if (REALPATH (entry->filename, fbuf) == NULL)
> > 		strncpy (fbuf, entry->filename, sizeof (fbuf));
> > is about I think.
> > 
> 
> You're right.  Then, the REALPATH call isn't needed anymore.
> It was there so we could get the filename extension of the dll
> in the case you mention.
> 
> I've added tests for symlinks.  Should I guard them for MinGW?  How?
> The symlink functionality comes from the host not from the target.
> Using msys or 'gcc -mno-cygwin' should work, no?
> 

Mingw-hosted ld.exe (whether in  msys env  or built with -mno-cygwin,
depends on MS msvcrt.dll,
which has not a clue about cygwin symlinks. Cygwin-hosted ld.exe depends
on cygwin1.dll so does recognize symlinks.

> Attached is the new patch, tested on i686-pc-cygwin. I'll try
> testing on MinGW later.
> 
That will be difficult.   MSVCRT.dll, hence mingw ld.exe,  has not a
clue about symlinks.  The deja-gnu tests expects ln -s to work.

In past, I have just accepted that if it doesn't break cygwin testsuite
results, its OK for mingw. (Many of the testcases--the bootstrap ones in
particular,  assume a POSIXish file system).

That is my long-winded way of saying  'nonetheless', thanks for the
patch  and thank Christopher for applying.
 
Danny

> Cheers,
> Pedro Alves
>