Re: Overriding LN_S

[Peter Rosin]

Den 2010-09-03 11:47 skrev Steffen Dettmer:
> On Fri, Sep 3, 2010 at 8:18 AM, Peter Rosin <address@hidden> wrote:
>> Den 2010-09-02 21:50 skrev Steffen Dettmer:
>>> On Thu, Sep 2, 2010 at 9:18 PM, Peter Rosin <address@hidden> wrote:
> [cp instead of ln -s]
>>>
>>> couldn't this break dependcies?
>>
>> I'm using MSVC to cross-compile from the Cygwin environment, so there
>> isn't any mixing of neither compilers nor dependencies.
> 
> OK, I just asked because we had issues here where constructions
> that created symlinks "once" but did no update them all the time,
> which is not needed for symlinks - but for copies it was needed of course.
> cp instead is pragmatic but might taste a bit like a
> "compatiblity hack" or so...

Of course it's a compatibility hack.

>>> Shouldn't "cygpath -w" be used itstead?
>>
>> "cygpath -w" is certainly a part of it, but has absolutely nothing to
>> do with the issue at hand. Libtool has the ability to create soft
>> links to the installed libltdl sources and the compiler then has to
>> be able to compile those sources "through" those soft links. If the
>> compiler doesn't interpret the soft links as soft links, but instead
>> as regular files, it has little chance of succeeding. So, I want to
>> tell libtool to make copies instead of soft links, as is done on
>> systems lacking soft links.
> 
> yes I see, of course.
> In general, I think it could be seen as part of the issue,
> because the system (cygwin) does have softlinks (and are better
> than copies), but MSVC and many other tools interpret paths
> differently and a translation in form of "cygpath -w" IMHO would
> be the clean natural approach (which also works for absolute
> paths and outputs those strange backslash-paths potentially
> needed by some windows tool etc).

Sorry, but there's just no way to make MSVC invoke cygpath when
it sees

#include "ltdl.h"

in some .c file and ltdl.h is a symlink.

Unless you propose an additional layer of preprocessing of source
files? Using (some variant of) LN_S="cp -p" seems like much less of
a "hack" to me.

> Interestingly, cygpath -w also seems to work for MSYS paths.

I highly doubt it'll do the right thing in all cases. My guess is
that you got "lucky" with whatever you tried, or that you have
some setup that makes it true for your case.

Cheers,
Peter