[Bug pch/65550] [4.8/4.9/5 Regression] ICE (segfault) with pch

["rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65550

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|ice-on-valid-code           |ice-on-invalid-code
           Priority|P3                          |P2
                 CC|                            |jsm28 at gcc dot gnu.org,
                   |                            |sje at gcc dot gnu.org
      Known to work|4.9.2                       |4.7.4
   Target Milestone|5.0                         |4.8.5
            Summary|[5 Regression] ICE          |[4.8/4.9/5 Regression] ICE
                   |(segfault) with pch         |(segfault) with pch
      Known to fail|                            |4.8.3, 4.9.2

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.

0xfb7742 crash_signal
        /space/rguenther/src/svn/trunk2/gcc/toplev.c:383
0x197315c linemap_add(line_maps*, lc_reason, unsigned int, char const*,
unsigned int)
        /space/rguenther/src/svn/trunk2/libcpp/line-map.c:303
0x1973b1e linemap_line_start(line_maps*, unsigned int, unsigned int)
        /space/rguenther/src/svn/trunk2/libcpp/line-map.c:566
0x1973d55 linemap_position_for_column(line_maps*, unsigned int)
        /space/rguenther/src/svn/trunk2/libcpp/line-map.c:611
0x1970744 _cpp_lex_direct
        /space/rguenther/src/svn/trunk2/libcpp/lex.c:2340
0x197029a _cpp_lex_token
        /space/rguenther/src/svn/trunk2/libcpp/lex.c:2176
0x197996f cpp_get_token_1
        /space/rguenther/src/svn/trunk2/libcpp/macro.c:2442
0x1979db7 cpp_get_token_with_location(cpp_reader*, unsigned int*)
        /space/rguenther/src/svn/trunk2/libcpp/macro.c:2628
0x9e2ea3 c_lex_with_flags(tree_node**, unsigned int*, unsigned char*, int)
        /space/rguenther/src/svn/trunk2/gcc/c-family/c-lex.c:408
#2  0x000000000197315d in linemap_add (set=0x10000fe000, reason=LC_RENAME, 
    sysp=0, to_file=0x24ad430 "../src/include.h", to_line=3)
    at /space/rguenther/src/svn/trunk2/libcpp/line-map.c:303
303       linemap_assert (!(set->depth == 0 && reason == LC_RENAME));
(gdb) p set->depth
$2 = 0


It's correctly initialized by PCH but re-set at

Hardware watchpoint 1: *$10

Old value = 1
New value = 0
linemap_add (set=0x10000fe000, reason=LC_LEAVE, sysp=0, to_file=0x0, to_line=0)
    at /space/rguenther/src/svn/trunk2/libcpp/line-map.c:311
311           return NULL;
(gdb) bt
#0  linemap_add (set=0x10000fe000, reason=LC_LEAVE, sysp=0, to_file=0x0, 
    to_line=0) at /space/rguenther/src/svn/trunk2/libcpp/line-map.c:311
#1  0x000000000195e29f in _cpp_do_file_change (pfile=0x249a670, 
    reason=LC_LEAVE, to_file=0x0, file_line=0, sysp=0)
    at /space/rguenther/src/svn/trunk2/libcpp/directives.c:1067
#2  0x0000000001961a5d in _cpp_pop_buffer (pfile=0x249a670)
    at /space/rguenther/src/svn/trunk2/libcpp/directives.c:2606
#3  0x00000000019704a7 in _cpp_get_fresh_line (pfile=0x249a670)
    at /space/rguenther/src/svn/trunk2/libcpp/lex.c:2255
#4  0x0000000001970587 in _cpp_lex_direct (pfile=0x249a670)
    at /space/rguenther/src/svn/trunk2/libcpp/lex.c:2302
#5  0x000000000197029b in _cpp_lex_token (pfile=0x249a670)
    at /space/rguenther/src/svn/trunk2/libcpp/lex.c:2176
#6  0x0000000001979970 in cpp_get_token_1 (pfile=0x249a670, 
    location=0x7fffffffdc58)
...
#10 0x000000000086afd3 in cp_parser_initial_pragma (first_token=0x7fffffffdc50)
    at /space/rguenther/src/svn/trunk2/gcc/cp/parser.c:32854
#11 0x000000000082eaf7 in cp_lexer_new_main ()
    at /space/rguenther/src/svn/trunk2/gcc/cp/parser.c:650
#12 0x0000000000831d4a in cp_parser_new ()
    at /space/rguenther/src/svn/trunk2/gcc/cp/parser.c:3484
#13 0x000000000086b787 in c_parse_file ()
    at /space/rguenther/src/svn/trunk2/gcc/cp/parser.c:33189



But note that this is an usupported use of PCH - it's included indirectly.
We should have rejected it but we don't.  Appearantly my fix for that doesn't
trigger:

2009-09-22  Richard Guenther  <rguenther@suse.de>

        PR pch/38987
        * files.c (pch_open_file): Disallow non-toplevel PCH inclusion.

because of the implicit-preinclude check.

(gdb) p pfile->all_files->implicit_preinclude
$30 = false
(gdb) p pfile->all_files->next_file->implicit_preinclude
$31 = true

I don't understand why we check two files here...

(gdb) p pfile->all_files->name
$33 = 0x24ad0c0 "include.h"
(gdb) p pfile->all_files->next_file->name
$34 = 0x24ac120 "stdc-predef.h"
(gdb) p pfile->all_files->next_file->next_file->name
$35 = 0x24e3be0 "../src/source.cpp"

note how the main file is "last".  My patch got two revisions - r192715 by
Joseph and r193709 by Steve.

I think the code should be changed to sth like

  /* If the file is not included as first include from either the toplevel
     file or the command-line it is not a valid use of PCH.  */
  for (_cpp_file *f = pfile->all_files; f; f = f->next_file)
    if (f->implicit_preinclude)
      continue;
    else if (f->main_file)
      break;
    else
      return false;

note the change triggering this bug is in GCC 4.8 already (but not in 4.7).


Index: libcpp/files.c
===================================================================
--- libcpp/files.c      (revision 221891)
+++ libcpp/files.c      (working copy)
@@ -291,11 +291,13 @@ pch_open_file (cpp_reader *pfile, _cpp_f

   /* If the file is not included as first include from either the toplevel
      file or the command-line it is not a valid use of PCH.  */
-  if (pfile->all_files
-      && pfile->all_files->next_file
-      && !(pfile->all_files->implicit_preinclude
-          || pfile->all_files->next_file->implicit_preinclude))
-    return false;
+  for (_cpp_file *f = pfile->all_files; f; f = f->next_file)
+    if (f->implicit_preinclude)
+      continue;
+    else if (f->main_file)
+      break;
+    else
+      return false;

   flen = strlen (path);
   len = flen + sizeof (extension);