Re: [Groff] need help resolving groff error

[Ralph Corderoy]

Hi charlie,

> we are more Windows knowledgeable.

Good, because I'm not and have investigated on Linux...  :-)

> > grops:<standard input>:4:fatal error: fdopen: Invalid argument

I just did a simple

    date | troff -Tps | grops >foo.ps

but with ltrace(1) monitoring grops so I could see what calls to
fdopen(3) occurred.  There was only one.

    mkstemp("/tmp/groffXXXXXX")  = 3
    fdopen(3, "w+")              = 0x2149ed0
    strlen("/tmp/groff4Jh8gU")   = 16

The fdopen() is probably src/libs/libgroff/tmpfile.cpp's from its
xtmpfile().

    FILE *xtmpfile(char **namep, const char *postfix_long,
        const char *postfix_short, int do_unlink)
    {
        char *templ = xtmptemplate(postfix_long, postfix_short);
        errno = 0;
        int fd = mkstemp(templ);
        if (fd < 0)
            fatal("cannot create temporary file: %1", strerror(errno));
        errno = 0;
        FILE *fp = fdopen(fd, FOPEN_RWB); // many callers of xtmpfile use 
binary I/O
        if (!fp)
            fatal("fdopen: %1", strerror(errno));

FOPEN_RWB is "w+" in my ltrace output but on Windows
src/include/nonposix.h makes me think it will be "wb+".  It looks like a
mismatch between the mode used to open the temporary file and the mode
that's then used to map a FILE* onto it with fdopen.

groff provides its own mkstemp() if the platform doesn't and that
open(2)s it with `O_RDWR | O_CREAT | O_EXCL'.  So either that's being
used and lacking the O_BINARY, that I think Windows defines, or Windows'
has a mkstemp that doesn't set `b' either.

Hope this helps those experienced with handling Windows's needs in the
groff source to cook up a solution.

-- 
Cheers, Ralph.
https://plus.google.com/115649437518703495227