Re: Text-mode issues

[Benjamin Riefenstahl <Benjamin dot Riefenstahl at epost dot de>]

Hi all,


I am consolidating my answers to todays post.


> Benjamin Riefenstahl wrote:
>> 1) The fonts directory must be on a binary mount.  This is documented
>>    and in the FAQs.  Given that the fonts directory is there only for
>>    XFree, this bug is minor.

Harold L Hunt II <huntharo@msu.edu> writes:
> The post-install scripts for the font packages were modified some
> time ago to automatically mount the fonts directory in bin mode.

I know.  I contributed to the discussion at the time, but as I wasn't
using XFree at the time, I didn't look into the real issue than.

> Did you have a specific problem where the fonts directory was not
> automatically mounted in bin mode, or are you just citing this as a
> weak area?

In my book an application that doesn't explicitly specify binary mode
when opening a binary file has a bug.  That's clearly how the issues
are defined for stdio (fopen()), and I believe for consistency it
should be handled in the same way for the POSIX APIs (open()).  I
personally regard everything else as a band-aid, and that certainly
goes for administrative solutions like binary mounts, those should be
*temporary* solutions IMO.

As I said, the administrative work-around works right now and other
programs usually don't use that directory, so I am not too concerned
here.

>> 2) The /tmp directory must be on a binary mount, otherwise
>>    international keyboards don't work.

Alexander Gottwald <alexander.gottwald@s1999.tu-chemnitz.de> writes:
> This is already fixed in CVS. But we've done no new release
> yet.

That is good to know, thanks.

> Have you tried not using -kb, or whatever else, and just trying
> Alexander's auto-detection code?  Or, does Alexander's
> auto-detection code still depend on /tmp?

I tried auto-detection first.  But I also tried the explicit option
and even installed XF86Config, they had all the same problem.

>> 3) The binary files in CVS are not marked as such (cvs admin -kb).

Igor Pechtchanski <pechtcha@cs.nyu.edu> writes:
> FYI, binary files in CVS should be marked with -kb even if they are
> extracted to binary mounts.  -kb only controls how the CVS keywords
> are expanded, so that if a binary file happens to contain a string
> like "$id$", it would not be expanded on checkout to something like
> "$Id: file.bin YYYY/MM/DD HH:mm:ss name " (thus effectively
> corrupting the binary).  If the binary files in the XFree86 tree
> aren't thus marked, it's a bug waiting to happen.

Thanks for the support.  Yes I know that this is the original idea of
-kb.  But that setting is also the recommended way of marking binary
files in the CVS for clients on OSs that have automatic text mode
translation.  While the bug is only potential for Unix, it's actual
for Windows in a CVS client implementation conforming with the OS
conventions, like cvsnt or Cygwin cvs on text mounts.

> [devel@xfree86.org] I am not sure if they have discussed this in the
> pass, though I guess that they have.

I haven't found it in the archives, I will post the question there.


Thanks for the feedback so far and
so long, benny