[tex-k] Size limitation for dvips output files

Karl Berry karl at freefriends.org
Fri Sep 17 02:14:31 CEST 2010


    dvips seems to have a size limit for output postscript files:
    http://bugs.debian.org/383781 at least on 32bit systems. Is there any
    chance to fix this?

Hi Hille -- the short answer is no (as far as I can tell).

Here's the long answer:

Despite your laudable attempts at recompiling with largefile support, I
continue to believe this restriction is about compiler options and not
anything in dvips specifically.  The error message "dvips: Problems with
file writing; probably disk full." is given upon any output error.
There's no magic constant in dvips preventing >2gb output.

I am not surprised that merely saying --enable-largefile in Build is not
sufficient.  I am not seeing any active largefile-related code in the
entire tree with the exception of one #define in
web2c/luatexdir/luafilesystem/src/lfs.c, hardly relevant here.  I guess
I should change the comments in the Build script (I hadn't realized this
earlier).

kpathsea/common.ac used to have AC_SYS_LARGEFILE, but we no longer use
that file.  There is no AC_SYS_LARGEFILE in any current configure that I
can see.  I also do not know if AC_SYS_LARGEFILE by itself is
sufficient, or if more changes to the source are necessary.

In any event, as stated in Build, the relevant declarations in TeX are
still inconsistent.  That's a different (and highly nontrivial) bug.
Furthermore, as mentioned in various places, there's a 32-bit limitation
in DVI format.  Yet another bug.  For these reasons, we will not be
compiling the whole tree with largefile support.

Furthermore, the infrastructure complications it would take to compile
some things (dvips) with largefile options and others (tex) without are
nontrivial.  We won't be doing that either, for the foreseeable future.

Meanwhile, the workaround for dvips that Tom gave ages ago, as reported
in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=383781 (dvips
foo.dvi -o - >foo.ps) seems reasonable to me.  The only reason to even
consider pursuing largefile at all, IMHO, is for
pdftex/luatex/xetex/dvipdfmx/whatever PDF output.

So, as far as I can see, it will take untold hours of effort to fix this
upstream.  Given that there are only a tiny number of developers and
many more urgent problems with real user impact to fix, I don't see
largefile happening without some new and very dedicated volunteer (you?)
coming forward to seriously sort out these issues.

If, for Debian, you feel like building dvips separately, and with
largefile support, go for it.  Patches most welcome.

So, overall, I'm afraid you may as well mark this wontfix.

Sorry,
karl


More information about the tex-k mailing list