[dvipdfmx] xdvipdfmx crashes with SIGSEGV for some Chinese documents

Mon May 20 21:25:30 CEST 2019

Hi,

I got an SIGSEGV issue around xdvipdfmx after upgrading texlive to
2019.2 on Arch Linux. Specifically, the following document cannot be
compiled with command `xelatex chinese.tex`:

\documentclass{article}
\usepackage{fontspec}
\usepackage{xeCJK}
\setCJKmainfont{AR PL UMing TW}
\begin{document}
中文內容
\end{document}

The error message is:

Error 139 (driver return code) generating output;

If I use "Noto Serif CJK TC" in \setCJKmainfont, the document compiles fine.

I managed to get a backtrace by splitting the original command into two:

$ xelatex -no-pdf chinese.tex
$ xdvipdfmx -o chinese.pdf < chinese.xdv

The gdb backtrace of the second command is:

#0 0x00005555555d9bf0 in add_ligature1_inverse_map
(cmap=cmap at entry=0x555555b70960,
used_chars=used_chars at entry=0x555555b5e010 "",
map_base=map_base at entry=0x555555ba7750,
map_sub=map_sub at entry=0x555555bc1f30,
num_glyphs=num_glyphs at entry=27123,
GIDToCIDMap=GIDToCIDMap at entry=0x555555bdc710, gid_1=40, idx=1,
data=0x555555b5dbf0) at ../../../texk/dvipdfm-x/tt_gsub.c:1891
#1 0x00005555555db2c6 in add_ligature1_inverse_map
(data=0x555555b5dbf0, idx=1, gid_1=<optimized out>,
GIDToCIDMap=0x555555bdc710, num_glyphs=27123, map_sub=0x555555bc1f30,
map_base=0x555555ba7750, used_chars=0x555555b5e010 "",
cmap=0x555555b70960) at ../../../texk/dvipdfm-x/tt_gsub.c:1962
#2 add_ToUnicode_ligature (GIDToCIDMap=0x555555bdc710,
num_glyphs=27123, map_sub=0x555555bc1f30, map_base=0x555555ba7750,
subtab=<optimized out>, used_chars=0x555555b5e010 "",
cmap=0x555555b70960) at ../../../texk/dvipdfm-x/tt_gsub.c:1962
#3 otl_gsub_add_ToUnicode (cmap=cmap at entry=0x555555b70960,
used_chars=used_chars at entry=0x555555b5e010 "",
map_base=map_base at entry=0x555555ba7750,
map_sub=map_sub at entry=0x555555bc1f30,
num_glyphs=num_glyphs at entry=27123,
GIDToCIDMap=GIDToCIDMap at entry=0x555555bdc710, sfont=0x555555b597d0) at
../../../texk/dvipdfm-x/tt_gsub.c:2017
#4 0x00005555555d42b6 in create_ToUnicode_cmap
(ttcmap=ttcmap at entry=0x555555b71960,
cmap_name=cmap_name at entry=0x555555b717f0 "FPXCHX+UMingTW-UTF16",
cmap_add=cmap_add at entry=0x0,
used_chars=used_chars at entry=0x555555b6ac30 "",
sfont=sfont at entry=0x555555b597d0) at
../../../texk/dvipdfm-x/tt_cmap.c:992
#5 0x00005555555d50e0 in otf_create_ToUnicode_stream
(font_name=0x555555b646e0 "/usr/share/fonts/TTF/uming.ttc",
ttc_index=ttc_index at entry=2, basefont=basefont at entry=0x555555b67120
"FPXCHX+UMingTW", used_chars=used_chars at entry=0x555555b6ac30 "") at
../../../texk/dvipdfm-x/tt_cmap.c:1174
#6 0x00005555555dd98c in Type0Font_attach_ToUnicode_stream
(font=0x555555b68810) at ../../../texk/dvipdfm-x/type0.c:328
#7 Type0Font_dofont (font=0x555555b68810) at ../../../texk/dvipdfm-x/type0.c:285
#8 Type0Font_cache_close () at ../../../texk/dvipdfm-x/type0.c:562
#9 0x00005555555a8529 in pdf_close_fonts () at
../../../texk/dvipdfm-x/pdffont.c:541
#10 0x00005555555a02fb in pdf_close_document () at
../../../texk/dvipdfm-x/pdfdoc.c:2584
#11 0x0000555555568e96 in main (argc=<optimized out>, argv=<optimized
out>) at ../../../texk/dvipdfm-x/dvipdfmx.c:1236

By the way, running the same command under valgrind gives a correct PDF file:

$ valgrind xdvipdfmx -o chinese.pdf < chinese.xdv
==28334== Memcheck, a memory error detector
==28334== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==28334== Using Valgrind-3.14.0 and LibVEX; rerun with -h for copyright info
==28334== Command: xdvipdfmx -o chinese.pdf
==28334==
stdin -> stdout
[1]==28334== Conditional jump or move depends on uninitialised value(s)
==28334== at 0x18DB9B: add_ligature1_inverse_map.part.7 (tt_gsub.c:1890)
==28334== by 0x18F2C5: add_ligature1_inverse_map (tt_gsub.c:1885)
==28334== by 0x18F2C5: add_ToUnicode_ligature (tt_gsub.c:1962)
==28334== by 0x18F2C5: otl_gsub_add_ToUnicode (tt_gsub.c:2017)
==28334== by 0x1882B5: create_ToUnicode_cmap (tt_cmap.c:992)
==28334== by 0x1890DF: otf_create_ToUnicode_stream (tt_cmap.c:1174)
==28334== by 0x19198B: Type0Font_attach_ToUnicode_stream (type0.c:232)
==28334== by 0x19198B: Type0Font_dofont (type0.c:285)
==28334== by 0x19198B: Type0Font_cache_close (type0.c:562)
==28334== by 0x15C528: pdf_close_fonts (pdffont.c:541)
==28334== by 0x1542FA: pdf_close_document (pdfdoc.c:2584)
==28334== by 0x11CE95: main (dvipdfmx.c:1236)
==28334==
==28334== Conditional jump or move depends on uninitialised value(s)
==28334== at 0x18DFE5: otl_gsub_release_ligature (tt_gsub.c:751)
==28334== by 0x18DFE5: otl_gsub_release.part.8 (tt_gsub.c:1443)
==28334== by 0x18F005: otl_gsub_release (tt_gsub.c:1420)
==28334== by 0x18F005: otl_gsub_add_ToUnicode (tt_gsub.c:2024)
==28334== by 0x1882B5: create_ToUnicode_cmap (tt_cmap.c:992)
==28334== by 0x1890DF: otf_create_ToUnicode_stream (tt_cmap.c:1174)
==28334== by 0x19198B: Type0Font_attach_ToUnicode_stream (type0.c:232)
==28334== by 0x19198B: Type0Font_dofont (type0.c:285)
==28334== by 0x19198B: Type0Font_cache_close (type0.c:562)
==28334== by 0x15C528: pdf_close_fonts (pdffont.c:541)
==28334== by 0x1542FA: pdf_close_document (pdfdoc.c:2584)
==28334== by 0x11CE95: main (dvipdfmx.c:1236)
==28334==
==28334== Conditional jump or move depends on uninitialised value(s)
==28334== at 0x4839961: free (vg_replace_malloc.c:530)
==28334== by 0x18E041: otl_gsub_release_ligature (tt_gsub.c:757)
==28334== by 0x18E041: otl_gsub_release.part.8 (tt_gsub.c:1443)
==28334== by 0x18F005: otl_gsub_release (tt_gsub.c:1420)
==28334== by 0x18F005: otl_gsub_add_ToUnicode (tt_gsub.c:2024)
==28334== by 0x1882B5: create_ToUnicode_cmap (tt_cmap.c:992)
==28334== by 0x1890DF: otf_create_ToUnicode_stream (tt_cmap.c:1174)
==28334== by 0x19198B: Type0Font_attach_ToUnicode_stream (type0.c:232)
==28334== by 0x19198B: Type0Font_dofont (type0.c:285)
==28334== by 0x19198B: Type0Font_cache_close (type0.c:562)
==28334== by 0x15C528: pdf_close_fonts (pdffont.c:541)
==28334== by 0x1542FA: pdf_close_document (pdfdoc.c:2584)
==28334== by 0x11CE95: main (dvipdfmx.c:1236)
==28334==

11877 bytes written
==28334==
==28334== HEAP SUMMARY:
==28334== in use at exit: 2,200,137 bytes in 82,764 blocks
==28334== total heap usage: 285,807 allocs, 203,043 frees, 16,483,280
bytes allocated
==28334==
==28334== LEAK SUMMARY:
==28334== definitely lost: 19,572 bytes in 733 blocks
==28334== indirectly lost: 3,996 bytes in 260 blocks
==28334== possibly lost: 0 bytes in 0 blocks
==28334== still reachable: 2,176,569 bytes in 81,771 blocks
==28334== suppressed: 0 bytes in 0 blocks
==28334== Rerun with --leak-check=full to see details of leaked memory
==28334==
==28334== For counts of detected and suppressed errors, rerun with: -v
==28334== Use --track-origins=yes to see where uninitialised values come from
==28334== ERROR SUMMARY: 168 errors from 3 contexts (suppressed: 0 from 0)

On Arch Linux, the xdvipdfmx command is provided by the package
texlive-bin, which is built from
https://github.com/TeX-Live/texlive-source/commit/74c2495978a4a84ffae10252c0fd244f1140228e.
Complete build script can be found at
https://git.archlinux.org/svntogit/packages.git/tree/trunk/PKGBUILD?h=packages/texlive-bin.
The fonts "AR PL UMing TW" and "Noto Serif CJK TC" are from
https://ftp.gnome.org/mirror/cdimage/snapshot/Debian/pool/main/t/ttf-arphic-uming/ttf-arphic-uming_0.2.20080216.1.orig.tar.gz
and https://github.com/googlefonts/noto-cjk/releases/tag/NotoSansV2.001,
respectively.

Thanks for any inputs!

Chih-Hsuan Yen