Some EPS file not handled correctly #302

Hi all,

I try to use Pillow in Py3.3 with EPS file.

First of all, all my test are made with this basic EPS file:

(download and rename the extension to EPS, Github does not allow eps directly and I hack it using Png extension :-p)

Many problems:

• This is a kind of EPS which begins by "0xC6D3D0C5". Currently, this kind of EPS cannot work in Python 3 because of this:

        s = fp.read(512)
        if s[:4] == b"%!PS":
            offset = 0
            fp.seek(0, 2)
            length = fp.tell()
        elif i32(s) == 0xC6D3D0C5:

And the PSfile implementation of read is:

    def read(self, count):
        return self.fp.read(count).decode('latin-1')

And "i32" raises TypeError in this case.

I think it is a bad idea to decode as a string here, moreover using latin-1 (at worst ascii). Sometimes EPS files can contain only ascii, but sometimes they contain binary data which will be not compatible with latin-1 conversion.

I have tried to fix converting all "str" to "bytes" (and the parsing is now OK), but I'm not sure this is compatible with Python 2 users...

Moreover, I'm not able to save it... The "length" read here:

        elif i32(s) == 0xC6D3D0C5:
            offset = i32(s[4:])
            length = i32(s[8:])
            fp.seek(offset)

is too small, and the loop inside "Ghoscript" function leads to a bad exit code while running "gs".... I don't known the EPS norm and I block at this point...

If someone is curious about correcting the EPS module :-)