Bugs item #315, was opened at 2005-03-21 16:53 You can respond by visiting: http://sarovar.org/tracker/?func=detail&atid=493&aid=315&group_id=106 Category: Fonts Group: v1.21a Status: Open Resolution: Fixed Priority: 5 Submitted By: Stan Swiercz (swiercz) Assigned to: Martin Schröder (oneiros) Summary: pdflatex on solaris font problems with acroread Initial Comment: I have compiled pdflatex 1.2.1a that comes with teTeX 3.0 on both linux and solaris. The linux version works properly but the solaris one does not. The last font included in the pdf output produces the error: Unable to extract the embedded font 'KTGBVU+CMR8'. Some characters may not display or print correctly. when viewed with acroread (version 5 or 6, windows, linux and solaris versions). Note that no error is seen when viewed with ghostscript and xpdf. The above error comes from the file: ------------------------ \documentclass{article} \newcommand{\ip}[2]{(#1, #2)} \begin{document} This is an example input file. Footnotes \footnote{This is an example of a footnote.} pose no problem. \LaTeX\ is good at typesetting mathematical formulas like \( x-3y + z = 7 \) or \( a_{1} > x^{2n} + y^{2n} > x' \) or \( \ip{A}{B} = \sum_{i} a_{i} b_{i} \). The spaces you type in a formula are ignored. Remember that a letter like $x$ % $ ... $ and \( ... \) are equivalent is a formula when it denotes a mathematical symbol, and it should be typed as one. \end{document} ----------------------- The sigma is produced correctly. If however I comment out the footnote, the sigma in the summation is no longer viewable and the error message is Unable to extract the embedded font 'ATSTIO+CMEX10'. Some characters may not display or print correctly. I attach a pdf files produced on solaris. Is this a know problem with a patch available? If you need further information please do not hesitate to contact me. Stan ----------------------------------------------------------------------
Comment By: Hartmut Henkel (hhenkel) Date: 2005-03-24 20:29
Message: Logged In: YES user_id=929 Hi Stan, if possible please test the attached patch to writet1.c. This now really initializes t1_save_offset to zero, and the save_offset() calls are only where really needed (there were too many, too early. They must never come before the first fb_putchar(), as only this does the alloc_array(fb,...) initialization. Also please do _not_ patch ptexmac.h (!), as this patch would cover otherwise maybe missing initializations. Leave ptexmac.h as in teTeX-3.0. I tried the writet1.c patch with arbitrary initial value of fb_ptr, and it seems to get it right. Thanks a lot for help! Regards, Hartmut ---------------------------------------------------------------------- Comment By: Hartmut Henkel (hhenkel) Date: 2005-03-24 16:48 Message: Logged In: YES user_id=929 who is pulling this bug into the "closed area"? :-) It seems to work but it still looks weird, maybe the final patch will be different... Regards, Hartmut ---------------------------------------------------------------------- Comment By: Stan Swiercz (swiercz) Date: 2005-03-24 16:18 Message: Logged In: YES user_id=2345 The ptexmac.h patch fixed it. Thank you! Stan ---------------------------------------------------------------------- Comment By: Hartmut Henkel (hhenkel) Date: 2005-03-24 16:04 Message: Logged In: YES user_id=929 once again, please try the patch for ptexmac.h. Regards, Hartmut ---------------------------------------------------------------------- Comment By: Stan Swiercz (swiercz) Date: 2005-03-24 16:01 Message: Logged In: YES user_id=2345 It does appear that t1_offset_value is at fault. I put in a pdftex_warn (something easy to add) in writet1.c and the first time it outputs I see Warning: pdflatex (file /nfs/encs/ArchDep/sun4.solaris8/pkg/teTeX-3.0/root/texm f/fonts/type1/bluesky/cm/cmr8.pfb): -268435455 t1_save_offset value All the other times the "warning" is displayed it returns 0. Stan ---------------------------------------------------------------------- Comment By: Hartmut Henkel (hhenkel) Date: 2005-03-24 15:52 Message: Logged In: YES user_id=929 thanks a lot, it was as expected :-( but the error _is_ around there, it's uninitialized fb_ptr i guess, just searching how to initialize it. Regards, Hartmut ---------------------------------------------------------------------- Comment By: Stan Swiercz (swiercz) Date: 2005-03-24 15:44 Message: Logged In: YES user_id=2345 I have applied the patch and found no difference. Stan ---------------------------------------------------------------------- Comment By: Hartmut Henkel (hhenkel) Date: 2005-03-24 15:29 Message: Logged In: YES user_id=929 maybe the infamous "Solaris compiler bug" is just an uninitialized variable t1_save_offset? It's unclear, as it seems that this could happen only if the font is not being included. Anyway: Could some Solaris freak try out the attached tiny patch to writet1.c? Regards, Hartmut ---------------------------------------------------------------------- Comment By: Thomas Esser (tetex) Date: 2005-03-24 01:35 Message: Logged In: YES user_id=1230 We have seen this before. pdftex list, posting by Adrian Lanz, Mon, 09 Aug 2004. There was no real outcome of that thread. Adrian wrote on 23 Sep 2004 to the texlive list: > On Tue, Sep 21, 2004 at 07:18:23PM -0500, Maksym Polyakov wrote: >> apparently there was some overflow in /Length1 -- it should be 810 > > Ah, now it sounds familiar. Adrian Lanz has reported the same to > pdftex@tug.org (09 Aug 2004, he uses SUN Solaris 9 / gcc 3.3.2). > Yes, I had exactly the same problem recently installing teTeX-beta 2.96.7.20040721. In my case, the symptom was that pdflatex run without error, but a wrong /Length1 parameter was inserted in the PDF output file, which evidently caused Acrobat Reader's error message (no problem, however, with xpdf or with the "dvips -Ppdf" routine). I think I solved the problem by changing the order and rules how shared libraries are found and handled on my Sun Sparc Solaris 2.9 system, and/or I may have re-installed some programs and libraries (ncurses, gettext, libiconv). I am sorry not being more specific on So, we have at least three people with this problem: Stan Swiercz, Maksym Polyakov, Adrian Lanz. I think that this is some king of compiler bug or something like that... Thomas ---------------------------------------------------------------------- Comment By: Martin Schröder (oneiros) Date: 2005-03-23 15:42 Message: Logged In: YES user_id=421 Yes, this is definitely a problem for some specific solaris installations which I can't reproduce. ---------------------------------------------------------------------- Comment By: Stan Swiercz (swiercz) Date: 2005-03-23 15:08 Message: Logged In: YES user_id=2345 I have done more debugging and I find that with a newer solaris 9 kernel the executable works fine. There is some patch to the OS that fixes this. This is something static as when I run this executable on a solaris machine running an older solaris9 kernel or solaris8 it works properly. Further digging I find that when I run the linker command on the older solaris 9 machine it produces a faulty executable BUT if I run "g++ -v" to see what the real linking command is and cut and paste it so as to run it instead the executable produces a pdf viewable with acroread. The basic g++ link command is c++ -o pdfetex pdfetexini.o pdfetex0.o pdfetex1.o pdfetex2.o pdfetex3.o pdfetexextra.o ../../libs/md5/md5.o pdftexdir/libpdf.a ../../libs/libpng/libpng.a ../../libs/zlib/libz.a ../../libs/xpdf/xpdf/libxpdf.a ../../libs/xpdf/goo/libGoo.a ../../libs/xpdf/fofi/libfofi.a -lsocket lib/lib.a ../kpathsea/.libs/libkpathsea.a -lm Instead of running this on older kernels I used /encs/pkg/gcc-3.3.2/root/lib/gcc-lib/sparc-sun-solaris2.8/3.3.2/collect2 -V -Y P,/usr/ccs/lib:/usr/lib -L /local/lib -L /encs/lib -Qy -o pdfetex /encs/pkg/gcc-3.3.2/root/lib/gcc-lib/sparc-sun-solaris2.8/3.3.2/crt1.o /encs/pkg/gcc-3.3.2/root/lib/gcc-lib/sparc-sun-solaris2.8/3.3.2/crti.o /usr/ccs/lib/values-Xa.o /encs/pkg/gcc-3.3.2/root/lib/gcc-lib/sparc-sun-solaris2.8/3.3.2/crtbegin.o -L/encs/bin/../lib/gcc-lib/sparc-sun-solaris2.8/3.3.2 -L/encs/bin/../lib/gcc-lib -L/encs/pkg/gcc-3.3.2/root/lib/gcc-lib/sparc-sun-solaris2.8/3.3.2 -L/usr/ccs/bin -L/usr/ccs/lib -L/encs/bin/../lib/gcc-lib/sparc-sun-solaris2.8/3.3.2/../../.. -L/encs/pkg/gcc-3.3.2/root/lib/gcc-lib/sparc-sun-solaris2.8/3.3.2/../../.. pdfetexini.o pdfetex0.o pdfetex1.o pdfetex2.o pdfetex3.o pdfetexextra.o ../../libs/md5/md5.o pdftexdir/libpdf.a ../../libs/libpng/libpng.a ../../libs/zlib/libz.a ../../libs/xpdf/xpdf/libxpdf.a ../../libs/xpdf/goo/libGoo.a ../../libs/xpdf/fofi/libfofi.a -lsocket lib/lib.a ../kpathsea/.libs/libkpathsea.a -lstdc++ -lm -lgcc_s -lgcc -lc -lgcc_s -lgcc -lc /encs/pkg/gcc-3.3.2/root/lib/gcc-lib/sparc-sun-solaris2.8/3.3.2/crtend.o /encs/pkg/gcc-3.3.2/root/lib/gcc-lib/sparc-sun-solaris2.8/3.3.2/crtn.o For the record /usr/ccs/lib/values-Xa.o is the same on both systems. I don't understrand... I also tracked down what the difference was between a pdf file that displays and does not display. The one that does display correctly has /Type /Page /Contents 49 0 R /Resources 47 0 R /MediaBox [0 0 612 792] /Parent 25 0 R
endobj 47 0 obj << /Font << /F8 12 0 R /F27 24 0 R /F11 31 0 R /F13 37 0 R /F7 18 0 R /F10 34 0 R >> /ProcSet [ /PDF /Text ] endobj 45 0 obj << /Length1 946 /Length2 3204 /Length3 532 /Length 3847 /Filter /FlateDecode
while the one that does not has /Type /Page /Contents 49 0 R /Resources 47 0 R /MediaBox [0 0 612 792] /Parent 25 0 R
endobj 47 0 obj << /Font << /F8 12 0 R /F27 24 0 R /F11 31 0 R /F13 37 0 R /F7 18 0 R /F10 34 0 R >> /ProcSet [ /PDF /Text ] endobj 45 0 obj << /Length1 268436401 /Length2 3204 /Length3 532 /Length 3848 /Filter /FlateDecode
the 946 has turned into 268436401! Hope this info helps. Stan ---------------------------------------------------------------------- Comment By: Martin Schröder (oneiros) Date: 2005-03-23 14:25 Message: Logged In: YES user_id=421 This seems to be a build problem on Solaris. Compiling the example on teTeX 3.0 on Linux produces a pdf that shows no problems in AR. A possible cause for the message by AR might be this extract from pdf_test_solaris.pdf: --------- 29 0 obj << /Length1 268436401 /Length2 3204 --------- Length1 is definitely wrong. ---------------------------------------------------------------------- Comment By: Martin Schröder (oneiros) Date: 2005-03-23 14:16 Message: Logged In: YES user_id=421 I can reproduce the problem with AR5 and AR7(latest "beta") on Linux. I can't reproduce it with gs 8.50 and Jaws. I suspect that this is a bug in AR. :-( ---------------------------------------------------------------------- You can respond by visiting: http://sarovar.org/tracker/?func=detail&atid=493&aid=315&group_id=106