ICE (segfault) in preprocessor

[gerhard]

gcc dies with segmentation violation on Intrepid and Karmic (didn't try others) with the following shell code; this is not an academic example but the maximally reduced version of my code:

echo -e '#include "b.h"\n#include "c.h"' >a.c; echo '# 1 "c.h" 1' >b.h; gcc -E a.c

ProblemType: Bug
Architecture: i386
DistroRelease: Ubuntu 8.10
Package: gcc 4:4.3.1-1ubuntu2
ProcEnviron:
 PATH=.:/home/username/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games
 LANG=en_US.UTF-8
 SHELL=/bin/bash
SourcePackage: gcc-defaults
Uname: Linux 2.6.27-16-generic i686

[Matthias Klose (doko)]

confirmed with 4.4 and 4.5

[Doko-v (doko-v)]

seen on any version back to 4.4

$ cat a.c
#include "b.h"
#include "c.h"
$ cat b.h
# 1 "c.h" 1
$ gcc -E a.c
# 1 "a.c"
# 1 "<built-in>"
# 1 "<command-line>"
# 1 "/usr/include/stdc-predef.h" 1 3 4
# 1 "<command-line>" 2
# 1 "a.c"
# 1 "b.h" 1
# 1 "c.h" 1
# 1 "b.h" 2
In file included from a.c:1:0:
b.h:1:0: internal compiler error: Segmentation fault
 # 1 "c.h" 1
 ^
Please submit a full bug report,
with preprocessed source if appropriate.

Program received signal SIGSEGV, Segmentation fault.
0x0000000000d990a8 in ?? ()
(gdb) bt
#0 0x0000000000d990a8 in ?? ()
#1 0x0000000000d997e8 in ?? ()
#2 0x0000000000d9a3ab in _cpp_stack_file ()
#3 0x0000000000d9a844 in _cpp_stack_include ()
#4 0x0000000000d92666 in ?? ()
#5 0x0000000000d931e8 in _cpp_handle_directive ()
#6 0x0000000000d9fa1d in _cpp_lex_token ()
#7 0x0000000000da42d0 in ?? ()
#8 0x00000000005d5af8 in preprocess_file(cpp_reader*) ()
#9 0x00000000005d4591 in c_common_init() ()
#10 0x000000000057faee in c_objc_common_init() ()
#11 0x00000000008b4fe7 in toplev_main(int, char**) ()
#12 0x00007ffff69e8ec5 in __libc_start_main (main=0x5526e0 <main>, argc=12, argv=0x7fffffffe5e8,
    init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7fffffffe5d8)
    at libc-start.c:287
#13 0x0000000000552915 in _start ()

[Rguenth (rguenth)]

I think we have a dup of this somewhere

[Matthias Klose (doko)]

seen with trunk 20210306

$ gcc-11 -E a.c
# 0 "a.c"
# 0 "<built-in>"
# 0 "<command-line>"
# 1 "/usr/include/stdc-predef.h" 1 3 4
# 0 "<command-line>" 2
# 1 "a.c"
# 1 "b.h" 1
# 1 "c.h" 1
# 1 "b.h" 2
In file included from a.c:1:
b.h:1: internal compiler error: Segmentation fault
    1 | # 1 "c.h" 1
      |
0xd041a0 crash_signal
        ../../src/gcc/toplev.c:327
0x11e616c open_file
        ../libcpp/../../src/libcpp/files.c:225
0x165a9ba read_file
        ../libcpp/../../src/libcpp/files.c:770
0x11e62dc _cpp_stack_file
        ../libcpp/../../src/libcpp/files.c:942
0x1684cd2 do_include_common
        ../libcpp/../../src/libcpp/directives.c:853
0x120a24f _cpp_handle_directive
        ../libcpp/../../src/libcpp/directives.c:541
0x1204fa4 _cpp_lex_token
        ../libcpp/../../src/libcpp/lex.c:2809
0x120927f cpp_get_token_1
        ../libcpp/../../src/libcpp/macro.c:2839
0x16841c3 cpp_get_token_with_location(cpp_reader*, unsigned int*)
        ../libcpp/../../src/libcpp/macro.c:3108
0x16841c3 scan_translation_unit
        ../../src/gcc/c-family/c-ppoutput.c:320
0x16841c3 preprocess_file(cpp_reader*)
        ../../src/gcc/c-family/c-ppoutput.c:102
0x167e491 c_common_init()
        ../../src/gcc/c-family/c-opts.c:1195
0x163fb1b lang_dependent_init
        ../../src/gcc/toplev.c:1889
0x163fb1b do_compile
        ../../src/gcc/toplev.c:2186
Please submit a full bug report,
with preprocessed source if appropriate.

[Cvs-commit (cvs-commit)]

The master branch has been updated by Lewis Hyatt <email address hidden>:

https://gcc.gnu.org/g:601dbf2a799f691688dfe78250b5bea2717b5b5e

commit r14-4185-g601dbf2a799f691688dfe78250b5bea2717b5b5e
Author: Lewis Hyatt <email address hidden>
Date: Fri Sep 15 13:31:51 2023 -0400

    libcpp: Fix ICE on #include after a line marker directive [PR61474]

    As noted in the PR, GCC will segfault if a file name is first seen in a
    linemarker directive, and then later seen in a normal #include. This is
    because the fake include process adds the file to the cache with a null PATH
    member. The normal #include finds this file in the cache and then attempts
    to use the null PATH. Resolve by adding the file to the cache with a unique
    starting directory, so that the fake entry will only be found by a
    subsequent fake include, not by a real one.

    libcpp/ChangeLog:

            PR preprocessor/61474
            * files.cc (_cpp_find_file): Set DONT_READ to TRUE for fake
            include files.
            (_cpp_fake_include): Pass a unique cpp_dir* address so
            the fake file will not be found when looked up for real.

    gcc/testsuite/ChangeLog:

            PR preprocessor/61474
            * c-c++-common/cpp/pr61474-2.h: New test.
            * c-c++-common/cpp/pr61474.c: New test.
            * c-c++-common/cpp/pr61474.h: New test.

[Lhyatt (lhyatt)]

Thanks for the testcase, this is fixed for GCC 14.